04a_MixMatch_extended.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-11T07:22:04.581817Z",
     "start_time": "2019-10-11T07:22:04.567550Z"
    },
    "hide_input": true
   },
   "outputs": [
    {
     "data": {
      "application/javascript": [
       "utils.load_extension('collapsible_headings/main')\n",
       "utils.load_extension('hide_input/main')\n",
       "utils.load_extension('autosavetime/main')\n",
       "utils.load_extension('execute_time/ExecuteTime')\n",
       "utils.load_extension('code_prettify/code_prettify')\n",
       "utils.load_extension('scroll_down/main')\n",
       "utils.load_extension('jupyter-js-widgets/extension')\n"
      ],
      "text/plain": [
       "<IPython.core.display.Javascript object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%%javascript\n",
    "utils.load_extension('collapsible_headings/main')\n",
    "utils.load_extension('hide_input/main')\n",
    "utils.load_extension('autosavetime/main')\n",
    "utils.load_extension('execute_time/ExecuteTime')\n",
    "utils.load_extension('code_prettify/code_prettify')\n",
    "utils.load_extension('scroll_down/main')\n",
    "utils.load_extension('jupyter-js-widgets/extension')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T12:34:27.975473Z",
     "start_time": "2019-10-15T12:34:27.970475Z"
    },
    "hide_input": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style>.container { width:100% !important; }</style>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from IPython.display import display, HTML\n",
    "display(HTML(\"<style>.container { width:100% !important; }</style>\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Ignacio Oguiza - email: oguiza@gmail.com"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true
   },
   "source": [
    "## A fastai/Pytorch implementation of MixMatch (Semi-supervised learning - SSL)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:49:41.512414Z",
     "start_time": "2019-10-15T17:49:41.507372Z"
    },
    "hidden": true
   },
   "source": [
    "Berthelot, D., Carlini, N., Goodfellow, I., Papernot, N., Oliver, A., & Raffel, C. (2019). **Mixmatch: A holistic approach to semi-supervised learning.** arXiv preprint arXiv:1905.02249. [paper](https://arxiv.org/pdf/1905.02249)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "I've created this notebook to show you how you can use MixMatch (a state-of-the-art semi-supervised learning technique) to leverage your unlabeled data.\n",
    "MixMatch works amazingly well in those cases where you have a relatively small amount of labeled data, but have access to unlabeled data.\n",
    "\n",
    "To build this notebook I've used fastai's excellent library, as well as an awesome [repo/notebook](https://github.com/noachr/MixMatch-fastai/blob/master/MixMatch%20Blog.ipynb) created by [Noah Rubinstein](https://github.com/noachr). Noah's notebook is a fantastic resource, the one I've seen to understand how MixMatch works. I highly recommend it!\n",
    "\n",
    "My main contributions in this notebook have been to:\n",
    "1. Simplify the code\n",
    "2. Make it easy to use"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true
   },
   "source": [
    "## Import libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:21:54.462946Z",
     "start_time": "2019-10-15T17:21:52.579359Z"
    },
    "hidden": true,
    "hide_input": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.0.58\n",
      "1.3.0\n"
     ]
    }
   ],
   "source": [
    "import fastai\n",
    "from fastai.vision import *\n",
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\", category=UserWarning, module=\"torch.nn.functional\")\n",
    "print(fastai.__version__)\n",
    "print(torch.__version__)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true
   },
   "source": [
    "## Prepare data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:21:56.388576Z",
     "start_time": "2019-10-15T17:21:56.379319Z"
    },
    "hidden": true
   },
   "outputs": [],
   "source": [
    "train_size = 500\n",
    "seed = 1234\n",
    "bs = 128\n",
    "path = untar_data(URLs.CIFAR)\n",
    "tfms = get_transforms()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "### Prepare simulated labeled/ unlabeled"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "☣️ In this notebook we'll simulate we have 500 labeled samples and 49500 unlabeled samples to train (the original CIFAR10 contains 50k labeled examples). To do that we'll use some random, stratified indices.\n",
    "\n",
    "The validation set will be the original one with 10k samples. You can skip this section when using your own data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:21:58.872259Z",
     "start_time": "2019-10-15T17:21:58.867361Z"
    },
    "hidden": true
   },
   "outputs": [],
   "source": [
    "def random_strat_splitter(y, train_size:int=1, seed:int=1):\n",
    "    from sklearn.model_selection import StratifiedShuffleSplit\n",
    "    sss = StratifiedShuffleSplit(n_splits=1, train_size=train_size, random_state=seed)\n",
    "    idx = list(sss.split(np.arange(len(y)), y))[0]\n",
    "    return idx[0],idx[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:22:00.493302Z",
     "start_time": "2019-10-15T17:21:59.770588Z"
    },
    "hidden": true
   },
   "outputs": [],
   "source": [
    "y_train = ImageList.from_folder(path / 'train').split_none().label_from_folder().y.items # 50k\n",
    "y_valid = ImageList.from_folder(path / 'test').split_none().label_from_folder().y.items  # 10k\n",
    "l_idx, u_idx = random_strat_splitter(y_train, train_size=train_size, seed=seed)  # train_size, 50k - train_size\n",
    "valid_idx, _ = random_strat_splitter(y_valid, train_size=train_size, seed=seed)  # 10k"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "### Prepare labeled databunch and list of unlabeled samples"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "1. Prepare your **labeled databunch** as you would normally do *.\n",
    "\n",
    "(*) [l_idx] is used to select a subsample of the original train dataset. You don't need this when using your own data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:22:03.504155Z",
     "start_time": "2019-10-15T17:22:02.589212Z"
    },
    "hidden": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ImageDataBunch;\n",
      "\n",
      "Train: LabelList (500 items)\n",
      "x: ImageList\n",
      "Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32)\n",
      "y: CategoryList\n",
      "deer,bird,truck,cat,deer\n",
      "Path: /home/oguizadl/.fastai/data/cifar10/train;\n",
      "\n",
      "Valid: LabelList (10000 items)\n",
      "x: ImageList\n",
      "Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32)\n",
      "y: CategoryList\n",
      "ship,ship,ship,ship,ship\n",
      "Path: /home/oguizadl/.fastai/data/cifar10/test;\n",
      "\n",
      "Test: None \n",
      "\n"
     ]
    }
   ],
   "source": [
    "data = (ItemLists('.',\n",
    "                  ImageList.from_folder(path / 'train')[l_idx],  # labeled train (in this case a subset of the original)\n",
    "                  ImageList.from_folder(path / 'test'))          # labeled valid\n",
    "        .label_from_folder()\n",
    "        .transform(tfms)\n",
    "        .databunch(bs=bs, val_bs=bs * 2)\n",
    "        .normalize(cifar_stats))\n",
    "print(data, '\\n')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "2. Now you need to create an **unlabeled ItemList** of the same type as the one used for labeled samples (in this case an ImageList, but it could be any type of ItemList). It includes all unlabeled samples you want to use *.\n",
    "\n",
    "\n",
    "(*) [u_idx] is used to select a subsample of the original train dataset. You don't need this when using your own data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:22:04.638124Z",
     "start_time": "2019-10-15T17:22:04.171661Z"
    },
    "hidden": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ImageList (49500 items)\n",
      "Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32)\n",
      "Path: /home/oguizadl/.fastai/data/cifar10/train \n",
      "\n"
     ]
    }
   ],
   "source": [
    "# unlabeled list\n",
    "ulist = ImageList.from_folder(path / 'train')[u_idx]\n",
    "print(ulist, '\\n')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:22:12.106686Z",
     "start_time": "2019-10-15T17:22:05.673572Z"
    },
    "hidden": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x1440 with 25 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "data.show_batch()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true
   },
   "source": [
    "## MixMatch code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:22:13.360879Z",
     "start_time": "2019-10-15T17:22:13.349727Z"
    },
    "hidden": true
   },
   "outputs": [],
   "source": [
    "from numbers import Integral\n",
    "\n",
    "class MultiTfmLabelList(LabelList):\n",
    "    def __init__(self, x:ItemList, y:ItemList, tfms:TfmList=None, tfm_y:bool=False, K=2, **kwargs):\n",
    "        \"K: number of transformed samples generated per item\"\n",
    "        self.x,self.y,self.tfm_y,self.K = x,y,tfm_y,K\n",
    "        self.y.x = x\n",
    "        self.item=None\n",
    "        self.transform(tfms, **kwargs)\n",
    "        \n",
    "    def __getitem__(self,idxs:Union[int, np.ndarray])->'LabelList':\n",
    "        \"return a single (x, y) if `idxs` is an integer or a new `LabelList` object if `idxs` is a range.\"\n",
    "        idxs = try_int(idxs)\n",
    "        if isinstance(idxs, Integral):\n",
    "            if self.item is None: x,y = self.x[idxs],self.y[idxs]\n",
    "            else:                 x,y = self.item   ,0\n",
    "            if self.tfms or self.tfmargs:\n",
    "                x = [x.apply_tfms(self.tfms, **self.tfmargs) for _ in range(self.K)]\n",
    "            if hasattr(self, 'tfms_y') and self.tfm_y and self.item is None:\n",
    "                y = y.apply_tfms(self.tfms_y, **{**self.tfmargs_y, 'do_resolve':False})\n",
    "            if y is None: y=0\n",
    "            return x,y\n",
    "        else: return self.new(self.x[idxs], self.y[idxs])\n",
    "\n",
    "def MultiCollate(batch):\n",
    "    batch = to_data(batch)\n",
    "    if isinstance(batch[0][0],list): batch = [[torch.stack(s[0]),s[1]] for s in batch]\n",
    "    return torch.utils.data.dataloader.default_collate(batch)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:22:14.645922Z",
     "start_time": "2019-10-15T17:22:14.636561Z"
    },
    "hidden": true
   },
   "outputs": [],
   "source": [
    "device = 'cuda' if torch.cuda.is_available() else 'cpu'\n",
    "\n",
    "def mixmatch(learn: Learner, ulist: ItemList, num_workers:int=None, \n",
    "             K: int = 2, T: float = .5, α: float = .75, λ: float = 100) -> Learner:\n",
    "\n",
    "    labeled_data = learn.data\n",
    "    if num_workers is None: num_workers = 1\n",
    "    labeled_data.train_dl.num_workers = num_workers\n",
    "    bs = labeled_data.train_dl.batch_size\n",
    "    tfms = [labeled_data.train_ds.tfms, labeled_data.valid_ds.tfms]\n",
    "    \n",
    "    ulist = ulist.split_none()\n",
    "    ulist.train._label_list = partial(MultiTfmLabelList, K=K)\n",
    "    train_ul = ulist.label_empty().train           # Train unlabeled Labelist\n",
    "    valid_ll = learn.data.label_list.valid         # Valid labeled Labelist\n",
    "    udata = (LabelLists('.', train_ul, valid_ll)\n",
    "             .transform(tfms)\n",
    "             .databunch(bs=min(bs, len(train_ul)),val_bs=min(bs * 2, len(valid_ll)),\n",
    "                        num_workers=num_workers,dl_tfms=learn.data.dl_tfms,device=device,\n",
    "                        collate_fn=MultiCollate)\n",
    "             .normalize(learn.data.stats))\n",
    "    learn.data = udata\n",
    "    learn.callback_fns.append(partial(MixMatchCallback, labeled_data=labeled_data, T=T, α=α, λ=λ))\n",
    "    return learn\n",
    "    \n",
    "Learner.mixmatch = mixmatch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:22:15.489913Z",
     "start_time": "2019-10-15T17:22:15.484193Z"
    },
    "hidden": true
   },
   "outputs": [],
   "source": [
    "def _mixup(x1, y1, x2, y2, α=.75):\n",
    "    β = np.random.beta(α, α)\n",
    "    β = max(β, 1 - β)\n",
    "    x = β * x1 + (1 - β) * x2\n",
    "    y = β * y1 + (1 - β) * y2\n",
    "    return x, y\n",
    "\n",
    "def sharpen(x, T=0.5):\n",
    "    p_target = x**(1. / T)\n",
    "    return p_target / p_target.sum(dim=1, keepdims=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:22:16.375643Z",
     "start_time": "2019-10-15T17:22:16.370956Z"
    },
    "hidden": true
   },
   "outputs": [],
   "source": [
    "def drop_cb_fn(learn, cb_name:str)->None:\n",
    "    cbs = []\n",
    "    for cb in learn.callback_fns:\n",
    "        if isinstance(cb, functools.partial): cbn = cb.func.__name__\n",
    "        else: cbn = cb.__name__\n",
    "        if cbn != cb_name: cbs.append(cb)\n",
    "    learn.callback_fns = cbs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:22:17.325951Z",
     "start_time": "2019-10-15T17:22:17.315660Z"
    },
    "hidden": true
   },
   "outputs": [],
   "source": [
    "class MatchMixLoss(Module):\n",
    "    \"Adapt the loss function `crit` to go with MatchMix.\"\n",
    "    \n",
    "    def __init__(self, crit=None, reduction='mean', λ=100):\n",
    "        super().__init__()\n",
    "        if crit is None: crit = nn.CrossEntropyLoss()\n",
    "        if hasattr(crit, 'reduction'): \n",
    "            self.crit = crit\n",
    "            self.old_red = crit.reduction\n",
    "            setattr(self.crit, 'reduction', 'none')\n",
    "        else: \n",
    "            self.crit = partial(crit, reduction='none')\n",
    "            self.old_crit = crit\n",
    "        self.reduction = reduction\n",
    "        self.λ = λ\n",
    "        \n",
    "    def forward(self, preds, target, bs=None):\n",
    "        \n",
    "        if bs is None: return F.cross_entropy(preds, target)\n",
    "\n",
    "        labeled_preds = torch.log_softmax(preds[:bs],dim=1)\n",
    "        Lx = -(labeled_preds * target[:bs]).sum(dim=1).mean()\n",
    "        self.Lx = Lx.item()\n",
    "\n",
    "        unlabeled_preds = torch.softmax(preds[bs:],dim=1)\n",
    "        Lu = F.mse_loss(unlabeled_preds,target[bs:])\n",
    "        self.Lu = (Lu * self.λ).item()\n",
    "        \n",
    "        return Lx + Lu * self.λ\n",
    "    \n",
    "    def get_old(self):\n",
    "        if hasattr(self, 'old_crit'):  return self.old_crit\n",
    "        elif hasattr(self, 'old_red'): \n",
    "            setattr(self.crit, 'reduction', self.old_red)\n",
    "            return self.crit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:22:19.914718Z",
     "start_time": "2019-10-15T17:22:19.894167Z"
    },
    "hidden": true
   },
   "outputs": [],
   "source": [
    "#export\n",
    "\n",
    "class MixMatchCallback(LearnerCallback):\n",
    "    _order = -20\n",
    "\n",
    "    def __init__(self,\n",
    "                 learn: Learner,\n",
    "                 labeled_data: DataBunch,\n",
    "                 T: float = .5,\n",
    "                 K: int = 2,\n",
    "                 α: float = .75,\n",
    "                 λ: float = 100):\n",
    "        super().__init__(learn)\n",
    "\n",
    "        self.learn, self.T, self.K, self.α, self.λ = learn, T, K, α, λ\n",
    "        self.labeled_dl = labeled_data.train_dl\n",
    "        self.n_classes = labeled_data.c\n",
    "        self.labeled_data = labeled_data\n",
    "        \n",
    "\n",
    "    def on_train_begin(self, n_epochs, **kwargs):\n",
    "        self.learn.loss_func = MatchMixLoss(crit=self.learn.loss_func, λ=self.λ)\n",
    "        self.ldliter = iter(self.labeled_dl)\n",
    "        self.smoothLx, self.smoothLu = SmoothenValue(0.98), SmoothenValue(0.98)\n",
    "        self.recorder.add_metric_names([\"train_Lx\", \"train_Lu*λ\"])\n",
    "        self.it = 0\n",
    "        print('labeled dataset     : {:13,} samples'.format(len(self.labeled_data.train_ds)))\n",
    "        print('unlabeled dataset   : {:13,} samples'.format(len(self.learn.data.train_ds)))\n",
    "        total_samples = n_epochs *len(self.learn.data.train_dl) *\\\n",
    "        self.learn.data.train_dl.batch_size * (self.K + 1)\n",
    "        print('total train samples : {:13,} samples'.format(total_samples))\n",
    "\n",
    "    def on_batch_begin(self, last_input, last_target, train, **kwargs):\n",
    "        if not train: return\n",
    "        try:\n",
    "            Xx, Xy = next(self.ldliter)             # Xx already augmented\n",
    "        except StopIteration:\n",
    "            self.ldliter = iter(self.labeled_dl)\n",
    "            Xx, Xy = next(self.ldliter)             # Xx already augmented\n",
    "\n",
    "        # LABELED\n",
    "        bs = len(Xx)\n",
    "        pb = torch.eye(self.n_classes)[Xy].to(device)\n",
    "\n",
    "        # UNLABELED\n",
    "        shape = list(last_input.size()[2:])\n",
    "        Ux = last_input.view([-1] + shape)           # Ux already augmented (K items)\n",
    "        with torch.no_grad():\n",
    "            Uy = sharpen(torch.softmax(torch.stack([\n",
    "                self.learn.model(last_input[:, i])\n",
    "                for i in range(last_input.shape[1])],dim=1),dim=2).mean(dim=1),T=self.T)\n",
    "        qb = Uy.repeat(1, 2).view((-1, Uy.size(-1)))\n",
    "\n",
    "        #MIX\n",
    "        Wx = torch.cat((Xx, Ux), dim=0)\n",
    "        Wy = torch.cat((pb, qb), dim=0)\n",
    "        shuffle = torch.randperm(Wx.shape[0])\n",
    "        mixed_input, mixed_target = _mixup(Wx, Wy, Wx[shuffle], Wy[shuffle], α=self.α)\n",
    "\n",
    "        return {\"last_input\": mixed_input, \"last_target\": (mixed_target, bs)}\n",
    "\n",
    "    def on_batch_end(self, train, **kwargs):\n",
    "        if not train: return\n",
    "        self.smoothLx.add_value(self.learn.loss_func.Lx)\n",
    "        self.smoothLu.add_value(self.learn.loss_func.Lu)\n",
    "        self.it += 1\n",
    "\n",
    "    def on_epoch_end(self, last_metrics, **kwargs):\n",
    "        return add_metrics(last_metrics, [self.smoothLx.smooth, self.smoothLu.smooth])\n",
    "    \n",
    "    def on_train_end(self, **kwargs):\n",
    "        \"\"\"At the end of training, loss_func and data are returned to their original values, \n",
    "        and this calleback is removed\"\"\"\n",
    "        self.learn.loss_func = self.learn.loss_func.get_old()\n",
    "        self.learn.data = self.labeled_data\n",
    "        drop_cb_fn(self.learn, 'MixMatchCallback')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true
   },
   "source": [
    "## Experiments"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true,
    "hidden": true
   },
   "source": [
    "### Supervised: 50k labeled samples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T14:44:40.648962Z",
     "start_time": "2019-10-15T14:44:38.304760Z"
    },
    "hidden": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ImageDataBunch;\n",
      "\n",
      "Train: LabelList (50000 items)\n",
      "x: ImageList\n",
      "Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32)\n",
      "y: CategoryList\n",
      "ship,ship,ship,ship,ship\n",
      "Path: /home/oguizadl/.fastai/data/cifar10/train;\n",
      "\n",
      "Valid: LabelList (10000 items)\n",
      "x: ImageList\n",
      "Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32)\n",
      "y: CategoryList\n",
      "ship,ship,ship,ship,ship\n",
      "Path: /home/oguizadl/.fastai/data/cifar10/test;\n",
      "\n",
      "Test: None \n",
      "\n"
     ]
    }
   ],
   "source": [
    "full_data = (ItemLists('.',\n",
    "                  ImageList.from_folder(path / 'train'),         # labeled train\n",
    "                  ImageList.from_folder(path / 'test'))          # labeled valid\n",
    "        .label_from_folder()\n",
    "        .transform(tfms)\n",
    "        .databunch(bs=bs, val_bs=bs * 2)\n",
    "        .normalize(cifar_stats))\n",
    "print(full_data, '\\n')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T16:02:34.452612Z",
     "start_time": "2019-10-15T14:44:40.651731Z"
    },
    "hidden": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>1.413102</td>\n",
       "      <td>1.322538</td>\n",
       "      <td>0.517300</td>\n",
       "      <td>01:44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>1.185652</td>\n",
       "      <td>1.235065</td>\n",
       "      <td>0.560000</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>1.077650</td>\n",
       "      <td>1.036387</td>\n",
       "      <td>0.626700</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.923953</td>\n",
       "      <td>0.888443</td>\n",
       "      <td>0.684000</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.833838</td>\n",
       "      <td>0.826511</td>\n",
       "      <td>0.710600</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.736363</td>\n",
       "      <td>0.733036</td>\n",
       "      <td>0.747200</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.671420</td>\n",
       "      <td>0.825750</td>\n",
       "      <td>0.728700</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.598679</td>\n",
       "      <td>0.681453</td>\n",
       "      <td>0.771800</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.546056</td>\n",
       "      <td>0.554246</td>\n",
       "      <td>0.811600</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.524312</td>\n",
       "      <td>0.631386</td>\n",
       "      <td>0.791700</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>0.491444</td>\n",
       "      <td>0.665037</td>\n",
       "      <td>0.785300</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>0.460563</td>\n",
       "      <td>0.472062</td>\n",
       "      <td>0.838600</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>0.437428</td>\n",
       "      <td>0.531470</td>\n",
       "      <td>0.822800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>0.434821</td>\n",
       "      <td>0.441585</td>\n",
       "      <td>0.849900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>0.405356</td>\n",
       "      <td>0.533483</td>\n",
       "      <td>0.824600</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>0.378884</td>\n",
       "      <td>0.408729</td>\n",
       "      <td>0.861500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>0.364254</td>\n",
       "      <td>0.376119</td>\n",
       "      <td>0.873800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>0.349018</td>\n",
       "      <td>0.380138</td>\n",
       "      <td>0.870600</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>0.333591</td>\n",
       "      <td>0.347152</td>\n",
       "      <td>0.881500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>0.312561</td>\n",
       "      <td>0.383316</td>\n",
       "      <td>0.873400</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>0.306224</td>\n",
       "      <td>0.407404</td>\n",
       "      <td>0.864900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>0.282770</td>\n",
       "      <td>0.378430</td>\n",
       "      <td>0.875900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>0.279758</td>\n",
       "      <td>0.353745</td>\n",
       "      <td>0.888400</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>0.261972</td>\n",
       "      <td>0.423823</td>\n",
       "      <td>0.866100</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>0.239697</td>\n",
       "      <td>0.334388</td>\n",
       "      <td>0.896100</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>0.228755</td>\n",
       "      <td>0.319845</td>\n",
       "      <td>0.899200</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>0.217304</td>\n",
       "      <td>0.320402</td>\n",
       "      <td>0.896200</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>0.206186</td>\n",
       "      <td>0.359357</td>\n",
       "      <td>0.888000</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>0.207623</td>\n",
       "      <td>0.288495</td>\n",
       "      <td>0.906800</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>0.191976</td>\n",
       "      <td>0.336646</td>\n",
       "      <td>0.892800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>0.185331</td>\n",
       "      <td>0.279169</td>\n",
       "      <td>0.912200</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>0.172976</td>\n",
       "      <td>0.298409</td>\n",
       "      <td>0.905700</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>0.175636</td>\n",
       "      <td>0.332772</td>\n",
       "      <td>0.899100</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>0.146367</td>\n",
       "      <td>0.280409</td>\n",
       "      <td>0.915900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>0.143775</td>\n",
       "      <td>0.298860</td>\n",
       "      <td>0.911500</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>0.143318</td>\n",
       "      <td>0.285142</td>\n",
       "      <td>0.912900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>36</td>\n",
       "      <td>0.129774</td>\n",
       "      <td>0.282496</td>\n",
       "      <td>0.915000</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>37</td>\n",
       "      <td>0.123828</td>\n",
       "      <td>0.261711</td>\n",
       "      <td>0.920500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>38</td>\n",
       "      <td>0.119421</td>\n",
       "      <td>0.254453</td>\n",
       "      <td>0.923200</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>39</td>\n",
       "      <td>0.115487</td>\n",
       "      <td>0.263524</td>\n",
       "      <td>0.920100</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>40</td>\n",
       "      <td>0.110395</td>\n",
       "      <td>0.297973</td>\n",
       "      <td>0.914100</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>41</td>\n",
       "      <td>0.107598</td>\n",
       "      <td>0.238006</td>\n",
       "      <td>0.927300</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>42</td>\n",
       "      <td>0.097759</td>\n",
       "      <td>0.268373</td>\n",
       "      <td>0.921900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>43</td>\n",
       "      <td>0.101271</td>\n",
       "      <td>0.269105</td>\n",
       "      <td>0.920500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>44</td>\n",
       "      <td>0.083784</td>\n",
       "      <td>0.279075</td>\n",
       "      <td>0.920200</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>45</td>\n",
       "      <td>0.096921</td>\n",
       "      <td>0.259707</td>\n",
       "      <td>0.926800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>46</td>\n",
       "      <td>0.085840</td>\n",
       "      <td>0.264013</td>\n",
       "      <td>0.925000</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>47</td>\n",
       "      <td>0.080873</td>\n",
       "      <td>0.311690</td>\n",
       "      <td>0.918700</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>48</td>\n",
       "      <td>0.080125</td>\n",
       "      <td>0.286986</td>\n",
       "      <td>0.920500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>49</td>\n",
       "      <td>0.071026</td>\n",
       "      <td>0.285051</td>\n",
       "      <td>0.925000</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>50</td>\n",
       "      <td>0.070023</td>\n",
       "      <td>0.257036</td>\n",
       "      <td>0.930600</td>\n",
       "      <td>00:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>51</td>\n",
       "      <td>0.068327</td>\n",
       "      <td>0.263028</td>\n",
       "      <td>0.929400</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>52</td>\n",
       "      <td>0.070450</td>\n",
       "      <td>0.263011</td>\n",
       "      <td>0.930200</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>53</td>\n",
       "      <td>0.066677</td>\n",
       "      <td>0.264080</td>\n",
       "      <td>0.929900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>54</td>\n",
       "      <td>0.058790</td>\n",
       "      <td>0.266175</td>\n",
       "      <td>0.930800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>55</td>\n",
       "      <td>0.051490</td>\n",
       "      <td>0.274422</td>\n",
       "      <td>0.928500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>56</td>\n",
       "      <td>0.058518</td>\n",
       "      <td>0.276713</td>\n",
       "      <td>0.930600</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>57</td>\n",
       "      <td>0.053288</td>\n",
       "      <td>0.252922</td>\n",
       "      <td>0.935500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>58</td>\n",
       "      <td>0.050697</td>\n",
       "      <td>0.280168</td>\n",
       "      <td>0.930500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>59</td>\n",
       "      <td>0.042558</td>\n",
       "      <td>0.251423</td>\n",
       "      <td>0.935800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>60</td>\n",
       "      <td>0.049446</td>\n",
       "      <td>0.272542</td>\n",
       "      <td>0.933100</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>61</td>\n",
       "      <td>0.038308</td>\n",
       "      <td>0.271401</td>\n",
       "      <td>0.933400</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>62</td>\n",
       "      <td>0.044663</td>\n",
       "      <td>0.264835</td>\n",
       "      <td>0.935100</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>63</td>\n",
       "      <td>0.034101</td>\n",
       "      <td>0.259427</td>\n",
       "      <td>0.934500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>64</td>\n",
       "      <td>0.035539</td>\n",
       "      <td>0.276778</td>\n",
       "      <td>0.933300</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>65</td>\n",
       "      <td>0.029356</td>\n",
       "      <td>0.283780</td>\n",
       "      <td>0.934100</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>66</td>\n",
       "      <td>0.031307</td>\n",
       "      <td>0.241937</td>\n",
       "      <td>0.941900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>67</td>\n",
       "      <td>0.021516</td>\n",
       "      <td>0.256339</td>\n",
       "      <td>0.939800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>68</td>\n",
       "      <td>0.021158</td>\n",
       "      <td>0.263525</td>\n",
       "      <td>0.937900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>69</td>\n",
       "      <td>0.021271</td>\n",
       "      <td>0.249267</td>\n",
       "      <td>0.940200</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>70</td>\n",
       "      <td>0.022885</td>\n",
       "      <td>0.249707</td>\n",
       "      <td>0.941300</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>71</td>\n",
       "      <td>0.019898</td>\n",
       "      <td>0.266895</td>\n",
       "      <td>0.939000</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>72</td>\n",
       "      <td>0.017840</td>\n",
       "      <td>0.267670</td>\n",
       "      <td>0.938900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>73</td>\n",
       "      <td>0.018212</td>\n",
       "      <td>0.278123</td>\n",
       "      <td>0.940500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>74</td>\n",
       "      <td>0.015287</td>\n",
       "      <td>0.271807</td>\n",
       "      <td>0.941700</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>75</td>\n",
       "      <td>0.015955</td>\n",
       "      <td>0.277990</td>\n",
       "      <td>0.939800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>76</td>\n",
       "      <td>0.015057</td>\n",
       "      <td>0.259213</td>\n",
       "      <td>0.943600</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>77</td>\n",
       "      <td>0.013993</td>\n",
       "      <td>0.266249</td>\n",
       "      <td>0.943600</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>78</td>\n",
       "      <td>0.011987</td>\n",
       "      <td>0.258395</td>\n",
       "      <td>0.944500</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>79</td>\n",
       "      <td>0.011516</td>\n",
       "      <td>0.256296</td>\n",
       "      <td>0.945900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>80</td>\n",
       "      <td>0.010671</td>\n",
       "      <td>0.274702</td>\n",
       "      <td>0.943900</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>81</td>\n",
       "      <td>0.010007</td>\n",
       "      <td>0.263114</td>\n",
       "      <td>0.947700</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>82</td>\n",
       "      <td>0.005572</td>\n",
       "      <td>0.254559</td>\n",
       "      <td>0.949700</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>83</td>\n",
       "      <td>0.006857</td>\n",
       "      <td>0.277788</td>\n",
       "      <td>0.946100</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>84</td>\n",
       "      <td>0.007465</td>\n",
       "      <td>0.259548</td>\n",
       "      <td>0.949600</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>85</td>\n",
       "      <td>0.006318</td>\n",
       "      <td>0.266455</td>\n",
       "      <td>0.948400</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>86</td>\n",
       "      <td>0.005086</td>\n",
       "      <td>0.270633</td>\n",
       "      <td>0.948200</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>87</td>\n",
       "      <td>0.003463</td>\n",
       "      <td>0.264249</td>\n",
       "      <td>0.949200</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>88</td>\n",
       "      <td>0.004764</td>\n",
       "      <td>0.267643</td>\n",
       "      <td>0.947800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>89</td>\n",
       "      <td>0.003334</td>\n",
       "      <td>0.261248</td>\n",
       "      <td>0.949800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>90</td>\n",
       "      <td>0.002978</td>\n",
       "      <td>0.269338</td>\n",
       "      <td>0.948200</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>91</td>\n",
       "      <td>0.003836</td>\n",
       "      <td>0.260643</td>\n",
       "      <td>0.949400</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>92</td>\n",
       "      <td>0.003030</td>\n",
       "      <td>0.265659</td>\n",
       "      <td>0.949700</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>93</td>\n",
       "      <td>0.002157</td>\n",
       "      <td>0.266552</td>\n",
       "      <td>0.949700</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>94</td>\n",
       "      <td>0.002545</td>\n",
       "      <td>0.266243</td>\n",
       "      <td>0.949100</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>95</td>\n",
       "      <td>0.003656</td>\n",
       "      <td>0.268339</td>\n",
       "      <td>0.949100</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>96</td>\n",
       "      <td>0.002977</td>\n",
       "      <td>0.268946</td>\n",
       "      <td>0.948800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>97</td>\n",
       "      <td>0.002872</td>\n",
       "      <td>0.264269</td>\n",
       "      <td>0.950400</td>\n",
       "      <td>00:47</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>98</td>\n",
       "      <td>0.004146</td>\n",
       "      <td>0.265309</td>\n",
       "      <td>0.949800</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>99</td>\n",
       "      <td>0.003319</td>\n",
       "      <td>0.264066</td>\n",
       "      <td>0.949700</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model = models.WideResNet(num_groups=3, N=4, num_classes=10, k=2, start_nf=32)\n",
    "learn = Learner(full_data, model, metrics=accuracy)\n",
    "learn.fit_one_cycle(100)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true,
    "hidden": true
   },
   "source": [
    "### Baseline - Supervised: 500 labeled samples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T12:35:54.569940Z",
     "start_time": "2019-10-15T12:35:54.551301Z"
    },
    "hidden": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ImageDataBunch;\n",
      "\n",
      "Train: LabelList (500 items)\n",
      "x: ImageList\n",
      "Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32)\n",
      "y: CategoryList\n",
      "deer,bird,truck,cat,deer\n",
      "Path: /home/oguizadl/.fastai/data/cifar10/train;\n",
      "\n",
      "Valid: LabelList (10000 items)\n",
      "x: ImageList\n",
      "Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32),Image (3, 32, 32)\n",
      "y: CategoryList\n",
      "ship,ship,ship,ship,ship\n",
      "Path: /home/oguizadl/.fastai/data/cifar10/test;\n",
      "\n",
      "Test: None\n"
     ]
    }
   ],
   "source": [
    "print(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T12:43:00.322469Z",
     "start_time": "2019-10-15T12:36:00.129958Z"
    },
    "hidden": true,
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>2.317151</td>\n",
       "      <td>2.290652</td>\n",
       "      <td>0.149500</td>\n",
       "      <td>00:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>2.290763</td>\n",
       "      <td>2.284799</td>\n",
       "      <td>0.169400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>2.260745</td>\n",
       "      <td>2.277761</td>\n",
       "      <td>0.175200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>2.233395</td>\n",
       "      <td>2.262488</td>\n",
       "      <td>0.193500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>2.208789</td>\n",
       "      <td>2.239619</td>\n",
       "      <td>0.209100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>2.183118</td>\n",
       "      <td>2.206290</td>\n",
       "      <td>0.234400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>2.151777</td>\n",
       "      <td>2.167754</td>\n",
       "      <td>0.260100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>2.122119</td>\n",
       "      <td>2.119564</td>\n",
       "      <td>0.273500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>2.088179</td>\n",
       "      <td>2.066700</td>\n",
       "      <td>0.284400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>2.049404</td>\n",
       "      <td>1.994711</td>\n",
       "      <td>0.298400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>2.007122</td>\n",
       "      <td>1.936763</td>\n",
       "      <td>0.298800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>1.966199</td>\n",
       "      <td>1.951179</td>\n",
       "      <td>0.296600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>1.920660</td>\n",
       "      <td>2.016528</td>\n",
       "      <td>0.289200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>1.883896</td>\n",
       "      <td>1.921566</td>\n",
       "      <td>0.339000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>1.849193</td>\n",
       "      <td>2.064118</td>\n",
       "      <td>0.342200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>1.807735</td>\n",
       "      <td>3.273612</td>\n",
       "      <td>0.262200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>1.766469</td>\n",
       "      <td>2.157545</td>\n",
       "      <td>0.323200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>1.723516</td>\n",
       "      <td>2.241818</td>\n",
       "      <td>0.323900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>1.689545</td>\n",
       "      <td>2.259183</td>\n",
       "      <td>0.340800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>1.645863</td>\n",
       "      <td>2.082204</td>\n",
       "      <td>0.352400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>1.617529</td>\n",
       "      <td>2.209355</td>\n",
       "      <td>0.356200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>1.585287</td>\n",
       "      <td>4.005031</td>\n",
       "      <td>0.292800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>1.554606</td>\n",
       "      <td>2.756733</td>\n",
       "      <td>0.324900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>1.521138</td>\n",
       "      <td>2.341560</td>\n",
       "      <td>0.307600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>1.495946</td>\n",
       "      <td>2.422877</td>\n",
       "      <td>0.328200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>1.462872</td>\n",
       "      <td>2.081654</td>\n",
       "      <td>0.358200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>1.426870</td>\n",
       "      <td>2.148562</td>\n",
       "      <td>0.345000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>1.396045</td>\n",
       "      <td>2.941402</td>\n",
       "      <td>0.300300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>1.366695</td>\n",
       "      <td>2.735512</td>\n",
       "      <td>0.291600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>1.336202</td>\n",
       "      <td>2.328615</td>\n",
       "      <td>0.387300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>1.307105</td>\n",
       "      <td>2.762975</td>\n",
       "      <td>0.335400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>1.281986</td>\n",
       "      <td>3.093845</td>\n",
       "      <td>0.332700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>1.259339</td>\n",
       "      <td>2.345172</td>\n",
       "      <td>0.356800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>1.232836</td>\n",
       "      <td>2.613732</td>\n",
       "      <td>0.350500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>1.203588</td>\n",
       "      <td>2.550699</td>\n",
       "      <td>0.362700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>1.181863</td>\n",
       "      <td>2.484174</td>\n",
       "      <td>0.319700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>36</td>\n",
       "      <td>1.161089</td>\n",
       "      <td>2.547053</td>\n",
       "      <td>0.358700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>37</td>\n",
       "      <td>1.133076</td>\n",
       "      <td>3.618252</td>\n",
       "      <td>0.331900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>38</td>\n",
       "      <td>1.102518</td>\n",
       "      <td>2.161794</td>\n",
       "      <td>0.413000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>39</td>\n",
       "      <td>1.069009</td>\n",
       "      <td>2.156173</td>\n",
       "      <td>0.396700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>40</td>\n",
       "      <td>1.044242</td>\n",
       "      <td>2.279373</td>\n",
       "      <td>0.394500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>41</td>\n",
       "      <td>1.013845</td>\n",
       "      <td>2.334417</td>\n",
       "      <td>0.393800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>42</td>\n",
       "      <td>0.986829</td>\n",
       "      <td>2.574715</td>\n",
       "      <td>0.379200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>43</td>\n",
       "      <td>0.955639</td>\n",
       "      <td>2.298761</td>\n",
       "      <td>0.410100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>44</td>\n",
       "      <td>0.925146</td>\n",
       "      <td>2.098819</td>\n",
       "      <td>0.423400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>45</td>\n",
       "      <td>0.896981</td>\n",
       "      <td>2.273616</td>\n",
       "      <td>0.411000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>46</td>\n",
       "      <td>0.867484</td>\n",
       "      <td>2.507767</td>\n",
       "      <td>0.403000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>47</td>\n",
       "      <td>0.838151</td>\n",
       "      <td>2.414421</td>\n",
       "      <td>0.390100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>48</td>\n",
       "      <td>0.811544</td>\n",
       "      <td>2.379093</td>\n",
       "      <td>0.413700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>49</td>\n",
       "      <td>0.780454</td>\n",
       "      <td>2.377183</td>\n",
       "      <td>0.423600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>50</td>\n",
       "      <td>0.756077</td>\n",
       "      <td>2.241371</td>\n",
       "      <td>0.437600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>51</td>\n",
       "      <td>0.727509</td>\n",
       "      <td>2.205027</td>\n",
       "      <td>0.443600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>52</td>\n",
       "      <td>0.700640</td>\n",
       "      <td>2.242137</td>\n",
       "      <td>0.433700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>53</td>\n",
       "      <td>0.676517</td>\n",
       "      <td>2.401623</td>\n",
       "      <td>0.421600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>54</td>\n",
       "      <td>0.650542</td>\n",
       "      <td>2.534793</td>\n",
       "      <td>0.416900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>55</td>\n",
       "      <td>0.632967</td>\n",
       "      <td>2.281995</td>\n",
       "      <td>0.436700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>56</td>\n",
       "      <td>0.609297</td>\n",
       "      <td>2.175525</td>\n",
       "      <td>0.446700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>57</td>\n",
       "      <td>0.586069</td>\n",
       "      <td>2.164386</td>\n",
       "      <td>0.454400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>58</td>\n",
       "      <td>0.564628</td>\n",
       "      <td>2.307937</td>\n",
       "      <td>0.435700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>59</td>\n",
       "      <td>0.540830</td>\n",
       "      <td>2.547261</td>\n",
       "      <td>0.417400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>60</td>\n",
       "      <td>0.521825</td>\n",
       "      <td>2.729192</td>\n",
       "      <td>0.416900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>61</td>\n",
       "      <td>0.500178</td>\n",
       "      <td>2.903299</td>\n",
       "      <td>0.402900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>62</td>\n",
       "      <td>0.479773</td>\n",
       "      <td>2.573395</td>\n",
       "      <td>0.420600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>63</td>\n",
       "      <td>0.459429</td>\n",
       "      <td>2.332593</td>\n",
       "      <td>0.445200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>64</td>\n",
       "      <td>0.440393</td>\n",
       "      <td>2.332423</td>\n",
       "      <td>0.454100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>65</td>\n",
       "      <td>0.423556</td>\n",
       "      <td>2.364201</td>\n",
       "      <td>0.450300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>66</td>\n",
       "      <td>0.407165</td>\n",
       "      <td>2.423923</td>\n",
       "      <td>0.444400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>67</td>\n",
       "      <td>0.390640</td>\n",
       "      <td>2.425673</td>\n",
       "      <td>0.446700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>68</td>\n",
       "      <td>0.374579</td>\n",
       "      <td>2.480034</td>\n",
       "      <td>0.447900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>69</td>\n",
       "      <td>0.360315</td>\n",
       "      <td>2.413413</td>\n",
       "      <td>0.450700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>70</td>\n",
       "      <td>0.343807</td>\n",
       "      <td>2.433969</td>\n",
       "      <td>0.445600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>71</td>\n",
       "      <td>0.330213</td>\n",
       "      <td>2.417513</td>\n",
       "      <td>0.449800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>72</td>\n",
       "      <td>0.314940</td>\n",
       "      <td>2.391640</td>\n",
       "      <td>0.458000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>73</td>\n",
       "      <td>0.301202</td>\n",
       "      <td>2.431542</td>\n",
       "      <td>0.458100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>74</td>\n",
       "      <td>0.290490</td>\n",
       "      <td>2.415626</td>\n",
       "      <td>0.459900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>75</td>\n",
       "      <td>0.277902</td>\n",
       "      <td>2.404630</td>\n",
       "      <td>0.457900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>76</td>\n",
       "      <td>0.265051</td>\n",
       "      <td>2.408396</td>\n",
       "      <td>0.458000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>77</td>\n",
       "      <td>0.252440</td>\n",
       "      <td>2.411658</td>\n",
       "      <td>0.457400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>78</td>\n",
       "      <td>0.241397</td>\n",
       "      <td>2.364168</td>\n",
       "      <td>0.463300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>79</td>\n",
       "      <td>0.231014</td>\n",
       "      <td>2.316371</td>\n",
       "      <td>0.467100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>80</td>\n",
       "      <td>0.220401</td>\n",
       "      <td>2.302724</td>\n",
       "      <td>0.465700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>81</td>\n",
       "      <td>0.209610</td>\n",
       "      <td>2.318193</td>\n",
       "      <td>0.466600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>82</td>\n",
       "      <td>0.200279</td>\n",
       "      <td>2.319210</td>\n",
       "      <td>0.467400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>83</td>\n",
       "      <td>0.190323</td>\n",
       "      <td>2.312550</td>\n",
       "      <td>0.467400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>84</td>\n",
       "      <td>0.181313</td>\n",
       "      <td>2.304590</td>\n",
       "      <td>0.470800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>85</td>\n",
       "      <td>0.172669</td>\n",
       "      <td>2.298469</td>\n",
       "      <td>0.474100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>86</td>\n",
       "      <td>0.163755</td>\n",
       "      <td>2.302902</td>\n",
       "      <td>0.473400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>87</td>\n",
       "      <td>0.155344</td>\n",
       "      <td>2.300956</td>\n",
       "      <td>0.474900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>88</td>\n",
       "      <td>0.147766</td>\n",
       "      <td>2.303031</td>\n",
       "      <td>0.474900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>89</td>\n",
       "      <td>0.141145</td>\n",
       "      <td>2.302419</td>\n",
       "      <td>0.475200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>90</td>\n",
       "      <td>0.134576</td>\n",
       "      <td>2.303180</td>\n",
       "      <td>0.474900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>91</td>\n",
       "      <td>0.128931</td>\n",
       "      <td>2.305237</td>\n",
       "      <td>0.476600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>92</td>\n",
       "      <td>0.123080</td>\n",
       "      <td>2.302846</td>\n",
       "      <td>0.477100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>93</td>\n",
       "      <td>0.117310</td>\n",
       "      <td>2.300483</td>\n",
       "      <td>0.477100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>94</td>\n",
       "      <td>0.111926</td>\n",
       "      <td>2.301567</td>\n",
       "      <td>0.476000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>95</td>\n",
       "      <td>0.107157</td>\n",
       "      <td>2.299855</td>\n",
       "      <td>0.477500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>96</td>\n",
       "      <td>0.101906</td>\n",
       "      <td>2.302311</td>\n",
       "      <td>0.476900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>97</td>\n",
       "      <td>0.097890</td>\n",
       "      <td>2.298620</td>\n",
       "      <td>0.475900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>98</td>\n",
       "      <td>0.093493</td>\n",
       "      <td>2.306107</td>\n",
       "      <td>0.477300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>99</td>\n",
       "      <td>0.089345</td>\n",
       "      <td>2.309372</td>\n",
       "      <td>0.476400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model = models.WideResNet(num_groups=3, N=4, num_classes=10, k=2, start_nf=32)\n",
    "learn = Learner(data, model, metrics=accuracy)\n",
    "learn.fit_one_cycle(100)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true,
    "hidden": true
   },
   "source": [
    "### Supervised: 500 labeled samples + Mixup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T12:53:21.448301Z",
     "start_time": "2019-10-15T12:46:26.425983Z"
    },
    "hidden": true,
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>2.333631</td>\n",
       "      <td>2.307531</td>\n",
       "      <td>0.135300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>2.308702</td>\n",
       "      <td>2.284551</td>\n",
       "      <td>0.159800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>2.286258</td>\n",
       "      <td>2.269721</td>\n",
       "      <td>0.173600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>2.268240</td>\n",
       "      <td>2.248985</td>\n",
       "      <td>0.177600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>2.249260</td>\n",
       "      <td>2.226807</td>\n",
       "      <td>0.204100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>2.235694</td>\n",
       "      <td>2.196989</td>\n",
       "      <td>0.218800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>2.218822</td>\n",
       "      <td>2.149207</td>\n",
       "      <td>0.222700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>2.198883</td>\n",
       "      <td>2.118525</td>\n",
       "      <td>0.227700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>2.177492</td>\n",
       "      <td>2.069287</td>\n",
       "      <td>0.253900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>2.158753</td>\n",
       "      <td>2.020361</td>\n",
       "      <td>0.270300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>2.138963</td>\n",
       "      <td>1.992471</td>\n",
       "      <td>0.281900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>2.117276</td>\n",
       "      <td>1.962980</td>\n",
       "      <td>0.290900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>2.095619</td>\n",
       "      <td>1.989998</td>\n",
       "      <td>0.280100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>2.076455</td>\n",
       "      <td>2.107301</td>\n",
       "      <td>0.260200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>2.055341</td>\n",
       "      <td>2.515268</td>\n",
       "      <td>0.247300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>2.041490</td>\n",
       "      <td>2.359928</td>\n",
       "      <td>0.288200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>2.023872</td>\n",
       "      <td>1.856569</td>\n",
       "      <td>0.348400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>2.008589</td>\n",
       "      <td>2.085950</td>\n",
       "      <td>0.284800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>1.990025</td>\n",
       "      <td>2.170291</td>\n",
       "      <td>0.315000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>1.974266</td>\n",
       "      <td>3.083651</td>\n",
       "      <td>0.269200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>1.958959</td>\n",
       "      <td>2.258923</td>\n",
       "      <td>0.309300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>1.941615</td>\n",
       "      <td>1.907755</td>\n",
       "      <td>0.346200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>1.925769</td>\n",
       "      <td>3.511420</td>\n",
       "      <td>0.216600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>1.912014</td>\n",
       "      <td>2.678666</td>\n",
       "      <td>0.289600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>1.899837</td>\n",
       "      <td>2.309119</td>\n",
       "      <td>0.286100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>1.888901</td>\n",
       "      <td>1.974136</td>\n",
       "      <td>0.348800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>1.876695</td>\n",
       "      <td>3.247090</td>\n",
       "      <td>0.201600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>1.865752</td>\n",
       "      <td>2.362855</td>\n",
       "      <td>0.286900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>1.854776</td>\n",
       "      <td>1.960396</td>\n",
       "      <td>0.343300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>1.839853</td>\n",
       "      <td>1.939738</td>\n",
       "      <td>0.343200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>1.831296</td>\n",
       "      <td>1.890805</td>\n",
       "      <td>0.364800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>1.820169</td>\n",
       "      <td>1.819465</td>\n",
       "      <td>0.391100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>1.806897</td>\n",
       "      <td>1.827476</td>\n",
       "      <td>0.394600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>1.790943</td>\n",
       "      <td>1.879700</td>\n",
       "      <td>0.378200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>1.782154</td>\n",
       "      <td>1.903733</td>\n",
       "      <td>0.365800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>1.771743</td>\n",
       "      <td>1.955925</td>\n",
       "      <td>0.364500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>36</td>\n",
       "      <td>1.758957</td>\n",
       "      <td>2.040055</td>\n",
       "      <td>0.333300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>37</td>\n",
       "      <td>1.747173</td>\n",
       "      <td>1.789615</td>\n",
       "      <td>0.394300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>38</td>\n",
       "      <td>1.735141</td>\n",
       "      <td>2.072751</td>\n",
       "      <td>0.343400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>39</td>\n",
       "      <td>1.719319</td>\n",
       "      <td>1.979599</td>\n",
       "      <td>0.364600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>40</td>\n",
       "      <td>1.708083</td>\n",
       "      <td>2.156963</td>\n",
       "      <td>0.332100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>41</td>\n",
       "      <td>1.693566</td>\n",
       "      <td>1.650541</td>\n",
       "      <td>0.426900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>42</td>\n",
       "      <td>1.677244</td>\n",
       "      <td>1.773416</td>\n",
       "      <td>0.413800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>43</td>\n",
       "      <td>1.663381</td>\n",
       "      <td>1.724866</td>\n",
       "      <td>0.424600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>44</td>\n",
       "      <td>1.648874</td>\n",
       "      <td>1.686053</td>\n",
       "      <td>0.422200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>45</td>\n",
       "      <td>1.634153</td>\n",
       "      <td>2.111795</td>\n",
       "      <td>0.346600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>46</td>\n",
       "      <td>1.620144</td>\n",
       "      <td>1.816069</td>\n",
       "      <td>0.397300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>47</td>\n",
       "      <td>1.610559</td>\n",
       "      <td>2.010760</td>\n",
       "      <td>0.379100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>48</td>\n",
       "      <td>1.597217</td>\n",
       "      <td>1.955747</td>\n",
       "      <td>0.384800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>49</td>\n",
       "      <td>1.583246</td>\n",
       "      <td>1.767821</td>\n",
       "      <td>0.411800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>50</td>\n",
       "      <td>1.568853</td>\n",
       "      <td>1.672256</td>\n",
       "      <td>0.438600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>51</td>\n",
       "      <td>1.554502</td>\n",
       "      <td>1.678773</td>\n",
       "      <td>0.436200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>52</td>\n",
       "      <td>1.538872</td>\n",
       "      <td>1.797112</td>\n",
       "      <td>0.435000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>53</td>\n",
       "      <td>1.526272</td>\n",
       "      <td>1.728930</td>\n",
       "      <td>0.451100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>54</td>\n",
       "      <td>1.509563</td>\n",
       "      <td>1.917263</td>\n",
       "      <td>0.428800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>55</td>\n",
       "      <td>1.499142</td>\n",
       "      <td>1.750555</td>\n",
       "      <td>0.432800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>56</td>\n",
       "      <td>1.486558</td>\n",
       "      <td>1.597007</td>\n",
       "      <td>0.457900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>57</td>\n",
       "      <td>1.470618</td>\n",
       "      <td>1.762445</td>\n",
       "      <td>0.412400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>58</td>\n",
       "      <td>1.457542</td>\n",
       "      <td>1.677205</td>\n",
       "      <td>0.444000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>59</td>\n",
       "      <td>1.444876</td>\n",
       "      <td>1.787596</td>\n",
       "      <td>0.421400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>60</td>\n",
       "      <td>1.428196</td>\n",
       "      <td>1.630615</td>\n",
       "      <td>0.459200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>61</td>\n",
       "      <td>1.418276</td>\n",
       "      <td>1.766402</td>\n",
       "      <td>0.432000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>62</td>\n",
       "      <td>1.407344</td>\n",
       "      <td>1.749127</td>\n",
       "      <td>0.439900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>63</td>\n",
       "      <td>1.395598</td>\n",
       "      <td>1.760665</td>\n",
       "      <td>0.445900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>64</td>\n",
       "      <td>1.380627</td>\n",
       "      <td>1.850919</td>\n",
       "      <td>0.436800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>65</td>\n",
       "      <td>1.370305</td>\n",
       "      <td>1.708106</td>\n",
       "      <td>0.462500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>66</td>\n",
       "      <td>1.359443</td>\n",
       "      <td>1.616789</td>\n",
       "      <td>0.474000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>67</td>\n",
       "      <td>1.349393</td>\n",
       "      <td>1.690714</td>\n",
       "      <td>0.459500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>68</td>\n",
       "      <td>1.332919</td>\n",
       "      <td>1.644398</td>\n",
       "      <td>0.468200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>69</td>\n",
       "      <td>1.321129</td>\n",
       "      <td>1.573894</td>\n",
       "      <td>0.488000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>70</td>\n",
       "      <td>1.312552</td>\n",
       "      <td>1.554680</td>\n",
       "      <td>0.490000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>71</td>\n",
       "      <td>1.301406</td>\n",
       "      <td>1.623856</td>\n",
       "      <td>0.478100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>72</td>\n",
       "      <td>1.292300</td>\n",
       "      <td>1.590700</td>\n",
       "      <td>0.484000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>73</td>\n",
       "      <td>1.279669</td>\n",
       "      <td>1.535379</td>\n",
       "      <td>0.493300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>74</td>\n",
       "      <td>1.268625</td>\n",
       "      <td>1.559630</td>\n",
       "      <td>0.489400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>75</td>\n",
       "      <td>1.259476</td>\n",
       "      <td>1.598149</td>\n",
       "      <td>0.478000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>76</td>\n",
       "      <td>1.251374</td>\n",
       "      <td>1.625041</td>\n",
       "      <td>0.471700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>77</td>\n",
       "      <td>1.242599</td>\n",
       "      <td>1.590638</td>\n",
       "      <td>0.484900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>78</td>\n",
       "      <td>1.232832</td>\n",
       "      <td>1.561637</td>\n",
       "      <td>0.495300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>79</td>\n",
       "      <td>1.224903</td>\n",
       "      <td>1.532731</td>\n",
       "      <td>0.503800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>80</td>\n",
       "      <td>1.213373</td>\n",
       "      <td>1.527813</td>\n",
       "      <td>0.505000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>81</td>\n",
       "      <td>1.205678</td>\n",
       "      <td>1.544053</td>\n",
       "      <td>0.501200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>82</td>\n",
       "      <td>1.196532</td>\n",
       "      <td>1.564783</td>\n",
       "      <td>0.498500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>83</td>\n",
       "      <td>1.190555</td>\n",
       "      <td>1.569077</td>\n",
       "      <td>0.496600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>84</td>\n",
       "      <td>1.183384</td>\n",
       "      <td>1.572523</td>\n",
       "      <td>0.497400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>85</td>\n",
       "      <td>1.175575</td>\n",
       "      <td>1.567878</td>\n",
       "      <td>0.495700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>86</td>\n",
       "      <td>1.167023</td>\n",
       "      <td>1.564913</td>\n",
       "      <td>0.493500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>87</td>\n",
       "      <td>1.159764</td>\n",
       "      <td>1.560256</td>\n",
       "      <td>0.494600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>88</td>\n",
       "      <td>1.155357</td>\n",
       "      <td>1.540667</td>\n",
       "      <td>0.500200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>89</td>\n",
       "      <td>1.147153</td>\n",
       "      <td>1.527202</td>\n",
       "      <td>0.503000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>90</td>\n",
       "      <td>1.140208</td>\n",
       "      <td>1.518140</td>\n",
       "      <td>0.504400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>91</td>\n",
       "      <td>1.131932</td>\n",
       "      <td>1.513264</td>\n",
       "      <td>0.506300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>92</td>\n",
       "      <td>1.123866</td>\n",
       "      <td>1.510397</td>\n",
       "      <td>0.506300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>93</td>\n",
       "      <td>1.119118</td>\n",
       "      <td>1.510090</td>\n",
       "      <td>0.505700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>94</td>\n",
       "      <td>1.110756</td>\n",
       "      <td>1.506557</td>\n",
       "      <td>0.506900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>95</td>\n",
       "      <td>1.103299</td>\n",
       "      <td>1.508141</td>\n",
       "      <td>0.507500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>96</td>\n",
       "      <td>1.098079</td>\n",
       "      <td>1.507560</td>\n",
       "      <td>0.507200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>97</td>\n",
       "      <td>1.092284</td>\n",
       "      <td>1.508244</td>\n",
       "      <td>0.507700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>98</td>\n",
       "      <td>1.087959</td>\n",
       "      <td>1.513663</td>\n",
       "      <td>0.505800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>99</td>\n",
       "      <td>1.088106</td>\n",
       "      <td>1.514042</td>\n",
       "      <td>0.507400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# mixup - alpha = .72 (as in MixMatch)\n",
    "model = models.WideResNet(num_groups=3, N=4, num_classes=10, k=2, start_nf=32)\n",
    "learn = Learner(data, model, metrics=accuracy).mixup(alpha=.75)\n",
    "learn.fit_one_cycle(100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T14:40:29.681088Z",
     "start_time": "2019-10-15T14:33:26.939598Z"
    },
    "hidden": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>2.306844</td>\n",
       "      <td>2.312384</td>\n",
       "      <td>0.117200</td>\n",
       "      <td>00:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>2.283394</td>\n",
       "      <td>2.282168</td>\n",
       "      <td>0.130400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>2.262537</td>\n",
       "      <td>2.269323</td>\n",
       "      <td>0.144000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>2.248402</td>\n",
       "      <td>2.249822</td>\n",
       "      <td>0.161400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>2.233027</td>\n",
       "      <td>2.231457</td>\n",
       "      <td>0.174900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>2.216703</td>\n",
       "      <td>2.215423</td>\n",
       "      <td>0.198000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>2.201614</td>\n",
       "      <td>2.167142</td>\n",
       "      <td>0.253200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>2.184015</td>\n",
       "      <td>2.137102</td>\n",
       "      <td>0.265300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>2.162226</td>\n",
       "      <td>2.113269</td>\n",
       "      <td>0.273300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>2.141590</td>\n",
       "      <td>2.045098</td>\n",
       "      <td>0.281900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>2.115216</td>\n",
       "      <td>1.987028</td>\n",
       "      <td>0.297300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>2.087970</td>\n",
       "      <td>1.939222</td>\n",
       "      <td>0.301400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>2.062558</td>\n",
       "      <td>1.904732</td>\n",
       "      <td>0.317300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>2.039284</td>\n",
       "      <td>2.489980</td>\n",
       "      <td>0.248100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>2.019179</td>\n",
       "      <td>1.832079</td>\n",
       "      <td>0.350900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>1.998942</td>\n",
       "      <td>2.169429</td>\n",
       "      <td>0.306500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>1.976281</td>\n",
       "      <td>2.468253</td>\n",
       "      <td>0.266100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>1.960330</td>\n",
       "      <td>2.115470</td>\n",
       "      <td>0.315900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>1.945699</td>\n",
       "      <td>2.640280</td>\n",
       "      <td>0.262700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>1.924405</td>\n",
       "      <td>2.181329</td>\n",
       "      <td>0.306500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>1.902891</td>\n",
       "      <td>3.371238</td>\n",
       "      <td>0.243000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>1.889060</td>\n",
       "      <td>2.012707</td>\n",
       "      <td>0.362100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>1.866529</td>\n",
       "      <td>2.218134</td>\n",
       "      <td>0.312700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>1.848043</td>\n",
       "      <td>2.077607</td>\n",
       "      <td>0.350400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>1.831807</td>\n",
       "      <td>2.301424</td>\n",
       "      <td>0.320100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>1.821997</td>\n",
       "      <td>2.102675</td>\n",
       "      <td>0.337500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>1.810305</td>\n",
       "      <td>2.353065</td>\n",
       "      <td>0.323100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>1.795450</td>\n",
       "      <td>1.854243</td>\n",
       "      <td>0.378700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>1.782399</td>\n",
       "      <td>1.827711</td>\n",
       "      <td>0.384700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>1.767123</td>\n",
       "      <td>2.575300</td>\n",
       "      <td>0.298300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>1.748523</td>\n",
       "      <td>2.200247</td>\n",
       "      <td>0.325800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>1.733665</td>\n",
       "      <td>2.750062</td>\n",
       "      <td>0.293500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>1.715129</td>\n",
       "      <td>2.494470</td>\n",
       "      <td>0.311300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>1.702963</td>\n",
       "      <td>2.332850</td>\n",
       "      <td>0.349700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>1.687341</td>\n",
       "      <td>2.030652</td>\n",
       "      <td>0.345200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>1.675564</td>\n",
       "      <td>2.151417</td>\n",
       "      <td>0.327500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>36</td>\n",
       "      <td>1.660543</td>\n",
       "      <td>2.194900</td>\n",
       "      <td>0.331500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>37</td>\n",
       "      <td>1.640926</td>\n",
       "      <td>1.960327</td>\n",
       "      <td>0.389200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>38</td>\n",
       "      <td>1.626658</td>\n",
       "      <td>2.257353</td>\n",
       "      <td>0.383300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>39</td>\n",
       "      <td>1.611517</td>\n",
       "      <td>2.064310</td>\n",
       "      <td>0.366300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>40</td>\n",
       "      <td>1.597120</td>\n",
       "      <td>2.101378</td>\n",
       "      <td>0.376700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>41</td>\n",
       "      <td>1.583183</td>\n",
       "      <td>1.992780</td>\n",
       "      <td>0.393000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>42</td>\n",
       "      <td>1.567101</td>\n",
       "      <td>1.935264</td>\n",
       "      <td>0.388700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>43</td>\n",
       "      <td>1.552546</td>\n",
       "      <td>2.214017</td>\n",
       "      <td>0.370400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>44</td>\n",
       "      <td>1.533111</td>\n",
       "      <td>1.767593</td>\n",
       "      <td>0.424600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>45</td>\n",
       "      <td>1.516650</td>\n",
       "      <td>1.714989</td>\n",
       "      <td>0.418900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>46</td>\n",
       "      <td>1.500194</td>\n",
       "      <td>1.731362</td>\n",
       "      <td>0.430100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>47</td>\n",
       "      <td>1.483652</td>\n",
       "      <td>1.867272</td>\n",
       "      <td>0.401700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>48</td>\n",
       "      <td>1.469437</td>\n",
       "      <td>1.976585</td>\n",
       "      <td>0.401100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>49</td>\n",
       "      <td>1.455963</td>\n",
       "      <td>2.195785</td>\n",
       "      <td>0.362100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>50</td>\n",
       "      <td>1.441306</td>\n",
       "      <td>1.895906</td>\n",
       "      <td>0.408100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>51</td>\n",
       "      <td>1.422623</td>\n",
       "      <td>1.805142</td>\n",
       "      <td>0.434000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>52</td>\n",
       "      <td>1.407558</td>\n",
       "      <td>1.784580</td>\n",
       "      <td>0.430300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>53</td>\n",
       "      <td>1.390559</td>\n",
       "      <td>1.887239</td>\n",
       "      <td>0.403800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>54</td>\n",
       "      <td>1.377866</td>\n",
       "      <td>1.761609</td>\n",
       "      <td>0.432100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>55</td>\n",
       "      <td>1.361356</td>\n",
       "      <td>1.804544</td>\n",
       "      <td>0.425500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>56</td>\n",
       "      <td>1.345819</td>\n",
       "      <td>1.770318</td>\n",
       "      <td>0.436400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>57</td>\n",
       "      <td>1.330326</td>\n",
       "      <td>2.255790</td>\n",
       "      <td>0.364800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>58</td>\n",
       "      <td>1.316970</td>\n",
       "      <td>1.841107</td>\n",
       "      <td>0.435900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>59</td>\n",
       "      <td>1.299077</td>\n",
       "      <td>1.713669</td>\n",
       "      <td>0.449600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>60</td>\n",
       "      <td>1.283295</td>\n",
       "      <td>1.753201</td>\n",
       "      <td>0.446000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>61</td>\n",
       "      <td>1.266771</td>\n",
       "      <td>1.634178</td>\n",
       "      <td>0.456000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>62</td>\n",
       "      <td>1.252599</td>\n",
       "      <td>1.726184</td>\n",
       "      <td>0.440100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>63</td>\n",
       "      <td>1.240029</td>\n",
       "      <td>1.864497</td>\n",
       "      <td>0.419900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>64</td>\n",
       "      <td>1.226501</td>\n",
       "      <td>1.768864</td>\n",
       "      <td>0.441600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>65</td>\n",
       "      <td>1.208549</td>\n",
       "      <td>1.765995</td>\n",
       "      <td>0.464300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>66</td>\n",
       "      <td>1.194143</td>\n",
       "      <td>1.769273</td>\n",
       "      <td>0.460800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>67</td>\n",
       "      <td>1.181584</td>\n",
       "      <td>1.782778</td>\n",
       "      <td>0.462000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>68</td>\n",
       "      <td>1.168132</td>\n",
       "      <td>1.705895</td>\n",
       "      <td>0.472900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>69</td>\n",
       "      <td>1.157284</td>\n",
       "      <td>1.681637</td>\n",
       "      <td>0.466900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>70</td>\n",
       "      <td>1.144379</td>\n",
       "      <td>1.647135</td>\n",
       "      <td>0.474400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>71</td>\n",
       "      <td>1.129912</td>\n",
       "      <td>1.640789</td>\n",
       "      <td>0.483000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>72</td>\n",
       "      <td>1.120052</td>\n",
       "      <td>1.679409</td>\n",
       "      <td>0.476400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>73</td>\n",
       "      <td>1.112916</td>\n",
       "      <td>1.740084</td>\n",
       "      <td>0.470600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>74</td>\n",
       "      <td>1.104211</td>\n",
       "      <td>1.729635</td>\n",
       "      <td>0.474000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>75</td>\n",
       "      <td>1.092027</td>\n",
       "      <td>1.679011</td>\n",
       "      <td>0.480200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>76</td>\n",
       "      <td>1.084220</td>\n",
       "      <td>1.608078</td>\n",
       "      <td>0.490300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>77</td>\n",
       "      <td>1.072190</td>\n",
       "      <td>1.570302</td>\n",
       "      <td>0.497400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>78</td>\n",
       "      <td>1.059982</td>\n",
       "      <td>1.573325</td>\n",
       "      <td>0.496800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>79</td>\n",
       "      <td>1.050613</td>\n",
       "      <td>1.598184</td>\n",
       "      <td>0.491500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>80</td>\n",
       "      <td>1.044954</td>\n",
       "      <td>1.619499</td>\n",
       "      <td>0.485300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>81</td>\n",
       "      <td>1.036214</td>\n",
       "      <td>1.635848</td>\n",
       "      <td>0.481900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>82</td>\n",
       "      <td>1.024375</td>\n",
       "      <td>1.643764</td>\n",
       "      <td>0.479600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>83</td>\n",
       "      <td>1.014442</td>\n",
       "      <td>1.638408</td>\n",
       "      <td>0.483400</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>84</td>\n",
       "      <td>1.008461</td>\n",
       "      <td>1.626792</td>\n",
       "      <td>0.486100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>85</td>\n",
       "      <td>0.997760</td>\n",
       "      <td>1.606529</td>\n",
       "      <td>0.489300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>86</td>\n",
       "      <td>0.992435</td>\n",
       "      <td>1.594517</td>\n",
       "      <td>0.490800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>87</td>\n",
       "      <td>0.981902</td>\n",
       "      <td>1.581428</td>\n",
       "      <td>0.493300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>88</td>\n",
       "      <td>0.973082</td>\n",
       "      <td>1.571297</td>\n",
       "      <td>0.495900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>89</td>\n",
       "      <td>0.965911</td>\n",
       "      <td>1.556016</td>\n",
       "      <td>0.501000</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>90</td>\n",
       "      <td>0.958501</td>\n",
       "      <td>1.551420</td>\n",
       "      <td>0.505500</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>91</td>\n",
       "      <td>0.953662</td>\n",
       "      <td>1.543250</td>\n",
       "      <td>0.506100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>92</td>\n",
       "      <td>0.950537</td>\n",
       "      <td>1.545262</td>\n",
       "      <td>0.505600</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>93</td>\n",
       "      <td>0.944616</td>\n",
       "      <td>1.543676</td>\n",
       "      <td>0.506700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>94</td>\n",
       "      <td>0.939731</td>\n",
       "      <td>1.540596</td>\n",
       "      <td>0.505300</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>95</td>\n",
       "      <td>0.935203</td>\n",
       "      <td>1.545880</td>\n",
       "      <td>0.503200</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>96</td>\n",
       "      <td>0.935898</td>\n",
       "      <td>1.545479</td>\n",
       "      <td>0.503900</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>97</td>\n",
       "      <td>0.934364</td>\n",
       "      <td>1.550746</td>\n",
       "      <td>0.502800</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>98</td>\n",
       "      <td>0.927467</td>\n",
       "      <td>1.544174</td>\n",
       "      <td>0.503700</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>99</td>\n",
       "      <td>0.922071</td>\n",
       "      <td>1.546697</td>\n",
       "      <td>0.503100</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# mixup - alpha = .5 (as in MixMatch's paper)\n",
    "model = models.WideResNet(num_groups=3, N=4, num_classes=10, k=2, start_nf=32)\n",
    "learn = Learner(data, model, metrics=accuracy).mixup(alpha=.5)\n",
    "learn.fit_one_cycle(100)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true,
    "hidden": true
   },
   "source": [
    "### Semi-supervised: 500 labeled + 49500 unlabeled samples using MixMatch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 143,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-14T19:27:19.158955Z",
     "start_time": "2019-10-14T19:27:18.610703Z"
    },
    "hidden": true,
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "model = models.WideResNet(num_groups=3, N=4, num_classes=10, k=2, start_nf=32)\n",
    "learn = Learner(data, model, metrics=accuracy).mixmatch(ulist, α=.75, λ=75)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 144,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T00:47:22.742651Z",
     "start_time": "2019-10-14T19:27:49.453405Z"
    },
    "hidden": true,
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "labeled ds  :    500\n",
      "unlabeled ds:  49500\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>train_Lx</th>\n",
       "      <th>train_Lu*λ</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>2.684314</td>\n",
       "      <td>2.892641</td>\n",
       "      <td>0.102800</td>\n",
       "      <td>2.186007</td>\n",
       "      <td>0.498305</td>\n",
       "      <td>03:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>2.018796</td>\n",
       "      <td>2.776649</td>\n",
       "      <td>0.140000</td>\n",
       "      <td>1.549589</td>\n",
       "      <td>0.469205</td>\n",
       "      <td>03:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>1.886331</td>\n",
       "      <td>2.621762</td>\n",
       "      <td>0.193400</td>\n",
       "      <td>1.299286</td>\n",
       "      <td>0.587042</td>\n",
       "      <td>03:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>1.711965</td>\n",
       "      <td>2.405672</td>\n",
       "      <td>0.270800</td>\n",
       "      <td>1.101404</td>\n",
       "      <td>0.610559</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>2.028985</td>\n",
       "      <td>1.715327</td>\n",
       "      <td>0.477000</td>\n",
       "      <td>1.157192</td>\n",
       "      <td>0.871791</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>1.918078</td>\n",
       "      <td>1.666925</td>\n",
       "      <td>0.498000</td>\n",
       "      <td>1.048883</td>\n",
       "      <td>0.869193</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>2.042164</td>\n",
       "      <td>1.700601</td>\n",
       "      <td>0.516100</td>\n",
       "      <td>1.089569</td>\n",
       "      <td>0.952594</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>2.086326</td>\n",
       "      <td>1.589027</td>\n",
       "      <td>0.523200</td>\n",
       "      <td>1.103491</td>\n",
       "      <td>0.982834</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>2.017304</td>\n",
       "      <td>1.538879</td>\n",
       "      <td>0.554200</td>\n",
       "      <td>1.068322</td>\n",
       "      <td>0.948981</td>\n",
       "      <td>03:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>1.939188</td>\n",
       "      <td>1.589867</td>\n",
       "      <td>0.556200</td>\n",
       "      <td>1.002264</td>\n",
       "      <td>0.936922</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>1.958657</td>\n",
       "      <td>1.341146</td>\n",
       "      <td>0.608300</td>\n",
       "      <td>1.021144</td>\n",
       "      <td>0.937512</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>1.925359</td>\n",
       "      <td>1.617297</td>\n",
       "      <td>0.571200</td>\n",
       "      <td>0.999485</td>\n",
       "      <td>0.925873</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>1.881601</td>\n",
       "      <td>1.517039</td>\n",
       "      <td>0.600900</td>\n",
       "      <td>0.952648</td>\n",
       "      <td>0.928952</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>1.622961</td>\n",
       "      <td>1.514951</td>\n",
       "      <td>0.594700</td>\n",
       "      <td>0.840904</td>\n",
       "      <td>0.782055</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>1.855460</td>\n",
       "      <td>1.306396</td>\n",
       "      <td>0.637700</td>\n",
       "      <td>0.963677</td>\n",
       "      <td>0.891781</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>1.897200</td>\n",
       "      <td>1.313821</td>\n",
       "      <td>0.636000</td>\n",
       "      <td>0.975011</td>\n",
       "      <td>0.922187</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>1.731713</td>\n",
       "      <td>1.246647</td>\n",
       "      <td>0.663100</td>\n",
       "      <td>0.891555</td>\n",
       "      <td>0.840156</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>1.764551</td>\n",
       "      <td>1.149960</td>\n",
       "      <td>0.681800</td>\n",
       "      <td>0.901507</td>\n",
       "      <td>0.863042</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>1.741365</td>\n",
       "      <td>1.129340</td>\n",
       "      <td>0.683800</td>\n",
       "      <td>0.871681</td>\n",
       "      <td>0.869682</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>1.713671</td>\n",
       "      <td>1.114353</td>\n",
       "      <td>0.693800</td>\n",
       "      <td>0.882869</td>\n",
       "      <td>0.830801</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>1.818257</td>\n",
       "      <td>1.297340</td>\n",
       "      <td>0.655900</td>\n",
       "      <td>0.926182</td>\n",
       "      <td>0.892074</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>1.739899</td>\n",
       "      <td>1.006029</td>\n",
       "      <td>0.709000</td>\n",
       "      <td>0.884440</td>\n",
       "      <td>0.855459</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>1.670190</td>\n",
       "      <td>0.983363</td>\n",
       "      <td>0.729600</td>\n",
       "      <td>0.870288</td>\n",
       "      <td>0.799902</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>1.660888</td>\n",
       "      <td>1.052651</td>\n",
       "      <td>0.711800</td>\n",
       "      <td>0.840769</td>\n",
       "      <td>0.820117</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>1.670573</td>\n",
       "      <td>1.003431</td>\n",
       "      <td>0.722600</td>\n",
       "      <td>0.861719</td>\n",
       "      <td>0.808853</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>1.604214</td>\n",
       "      <td>1.023640</td>\n",
       "      <td>0.728200</td>\n",
       "      <td>0.821762</td>\n",
       "      <td>0.782450</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>1.891168</td>\n",
       "      <td>0.868633</td>\n",
       "      <td>0.749400</td>\n",
       "      <td>0.972888</td>\n",
       "      <td>0.918279</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>1.609051</td>\n",
       "      <td>1.011954</td>\n",
       "      <td>0.725700</td>\n",
       "      <td>0.819728</td>\n",
       "      <td>0.789321</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>1.648536</td>\n",
       "      <td>0.975326</td>\n",
       "      <td>0.743400</td>\n",
       "      <td>0.834879</td>\n",
       "      <td>0.813654</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>1.650765</td>\n",
       "      <td>0.945522</td>\n",
       "      <td>0.753800</td>\n",
       "      <td>0.833362</td>\n",
       "      <td>0.817402</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>1.581052</td>\n",
       "      <td>0.891314</td>\n",
       "      <td>0.747200</td>\n",
       "      <td>0.796803</td>\n",
       "      <td>0.784247</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>1.703347</td>\n",
       "      <td>0.813212</td>\n",
       "      <td>0.779000</td>\n",
       "      <td>0.880589</td>\n",
       "      <td>0.822756</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>1.600958</td>\n",
       "      <td>0.877771</td>\n",
       "      <td>0.755000</td>\n",
       "      <td>0.805102</td>\n",
       "      <td>0.795854</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>1.673129</td>\n",
       "      <td>0.839794</td>\n",
       "      <td>0.770000</td>\n",
       "      <td>0.850755</td>\n",
       "      <td>0.822373</td>\n",
       "      <td>03:13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>1.495979</td>\n",
       "      <td>0.856562</td>\n",
       "      <td>0.771000</td>\n",
       "      <td>0.751112</td>\n",
       "      <td>0.744866</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>1.555366</td>\n",
       "      <td>0.820187</td>\n",
       "      <td>0.779500</td>\n",
       "      <td>0.800991</td>\n",
       "      <td>0.754374</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>36</td>\n",
       "      <td>1.394809</td>\n",
       "      <td>0.938598</td>\n",
       "      <td>0.750000</td>\n",
       "      <td>0.723686</td>\n",
       "      <td>0.671121</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>37</td>\n",
       "      <td>1.728298</td>\n",
       "      <td>0.770861</td>\n",
       "      <td>0.786300</td>\n",
       "      <td>0.872480</td>\n",
       "      <td>0.855816</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>38</td>\n",
       "      <td>1.437784</td>\n",
       "      <td>0.794928</td>\n",
       "      <td>0.782200</td>\n",
       "      <td>0.746783</td>\n",
       "      <td>0.691000</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>39</td>\n",
       "      <td>1.568687</td>\n",
       "      <td>0.768634</td>\n",
       "      <td>0.790200</td>\n",
       "      <td>0.804053</td>\n",
       "      <td>0.764633</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>40</td>\n",
       "      <td>1.457061</td>\n",
       "      <td>0.781324</td>\n",
       "      <td>0.789400</td>\n",
       "      <td>0.756982</td>\n",
       "      <td>0.700077</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>41</td>\n",
       "      <td>1.437266</td>\n",
       "      <td>0.790995</td>\n",
       "      <td>0.781900</td>\n",
       "      <td>0.762187</td>\n",
       "      <td>0.675078</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>42</td>\n",
       "      <td>1.536097</td>\n",
       "      <td>0.722292</td>\n",
       "      <td>0.806100</td>\n",
       "      <td>0.791564</td>\n",
       "      <td>0.744531</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>43</td>\n",
       "      <td>1.428884</td>\n",
       "      <td>0.735917</td>\n",
       "      <td>0.805500</td>\n",
       "      <td>0.732246</td>\n",
       "      <td>0.696637</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>44</td>\n",
       "      <td>1.515188</td>\n",
       "      <td>0.765575</td>\n",
       "      <td>0.799100</td>\n",
       "      <td>0.784523</td>\n",
       "      <td>0.730663</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>45</td>\n",
       "      <td>1.417510</td>\n",
       "      <td>0.794199</td>\n",
       "      <td>0.797600</td>\n",
       "      <td>0.735782</td>\n",
       "      <td>0.681726</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>46</td>\n",
       "      <td>1.655026</td>\n",
       "      <td>0.720324</td>\n",
       "      <td>0.799400</td>\n",
       "      <td>0.834958</td>\n",
       "      <td>0.820067</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>47</td>\n",
       "      <td>1.478441</td>\n",
       "      <td>0.740623</td>\n",
       "      <td>0.800600</td>\n",
       "      <td>0.749685</td>\n",
       "      <td>0.728755</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>48</td>\n",
       "      <td>1.343099</td>\n",
       "      <td>0.730294</td>\n",
       "      <td>0.806800</td>\n",
       "      <td>0.702006</td>\n",
       "      <td>0.641092</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>49</td>\n",
       "      <td>1.518929</td>\n",
       "      <td>0.739519</td>\n",
       "      <td>0.802100</td>\n",
       "      <td>0.790308</td>\n",
       "      <td>0.728619</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>50</td>\n",
       "      <td>1.451367</td>\n",
       "      <td>0.735893</td>\n",
       "      <td>0.802100</td>\n",
       "      <td>0.759775</td>\n",
       "      <td>0.691591</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>51</td>\n",
       "      <td>1.461456</td>\n",
       "      <td>0.715765</td>\n",
       "      <td>0.813000</td>\n",
       "      <td>0.771853</td>\n",
       "      <td>0.689602</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>52</td>\n",
       "      <td>1.473971</td>\n",
       "      <td>0.721928</td>\n",
       "      <td>0.813700</td>\n",
       "      <td>0.764014</td>\n",
       "      <td>0.709955</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>53</td>\n",
       "      <td>1.452259</td>\n",
       "      <td>0.712356</td>\n",
       "      <td>0.816200</td>\n",
       "      <td>0.745284</td>\n",
       "      <td>0.706973</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>54</td>\n",
       "      <td>1.457291</td>\n",
       "      <td>0.699870</td>\n",
       "      <td>0.810500</td>\n",
       "      <td>0.758741</td>\n",
       "      <td>0.698549</td>\n",
       "      <td>03:13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>55</td>\n",
       "      <td>1.437048</td>\n",
       "      <td>0.670193</td>\n",
       "      <td>0.819700</td>\n",
       "      <td>0.747665</td>\n",
       "      <td>0.689382</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>56</td>\n",
       "      <td>1.324067</td>\n",
       "      <td>0.645602</td>\n",
       "      <td>0.827600</td>\n",
       "      <td>0.687135</td>\n",
       "      <td>0.636931</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>57</td>\n",
       "      <td>1.410022</td>\n",
       "      <td>0.656911</td>\n",
       "      <td>0.822900</td>\n",
       "      <td>0.734535</td>\n",
       "      <td>0.675486</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>58</td>\n",
       "      <td>1.356676</td>\n",
       "      <td>0.669916</td>\n",
       "      <td>0.823700</td>\n",
       "      <td>0.713756</td>\n",
       "      <td>0.642919</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>59</td>\n",
       "      <td>1.374593</td>\n",
       "      <td>0.689691</td>\n",
       "      <td>0.829000</td>\n",
       "      <td>0.713545</td>\n",
       "      <td>0.661047</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>60</td>\n",
       "      <td>1.273278</td>\n",
       "      <td>0.672429</td>\n",
       "      <td>0.827600</td>\n",
       "      <td>0.680434</td>\n",
       "      <td>0.592842</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>61</td>\n",
       "      <td>1.249914</td>\n",
       "      <td>0.627292</td>\n",
       "      <td>0.839000</td>\n",
       "      <td>0.687475</td>\n",
       "      <td>0.562438</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>62</td>\n",
       "      <td>1.292115</td>\n",
       "      <td>0.637413</td>\n",
       "      <td>0.834700</td>\n",
       "      <td>0.685753</td>\n",
       "      <td>0.606361</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>63</td>\n",
       "      <td>1.371459</td>\n",
       "      <td>0.632097</td>\n",
       "      <td>0.838800</td>\n",
       "      <td>0.728589</td>\n",
       "      <td>0.642869</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>64</td>\n",
       "      <td>1.379985</td>\n",
       "      <td>0.613133</td>\n",
       "      <td>0.838100</td>\n",
       "      <td>0.735508</td>\n",
       "      <td>0.644476</td>\n",
       "      <td>03:13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>65</td>\n",
       "      <td>1.338347</td>\n",
       "      <td>0.642171</td>\n",
       "      <td>0.838100</td>\n",
       "      <td>0.715851</td>\n",
       "      <td>0.622495</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>66</td>\n",
       "      <td>1.222916</td>\n",
       "      <td>0.638388</td>\n",
       "      <td>0.835500</td>\n",
       "      <td>0.670907</td>\n",
       "      <td>0.552008</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>67</td>\n",
       "      <td>1.220347</td>\n",
       "      <td>0.598997</td>\n",
       "      <td>0.848800</td>\n",
       "      <td>0.671459</td>\n",
       "      <td>0.548887</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>68</td>\n",
       "      <td>1.248978</td>\n",
       "      <td>0.613540</td>\n",
       "      <td>0.842700</td>\n",
       "      <td>0.685824</td>\n",
       "      <td>0.563153</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>69</td>\n",
       "      <td>1.293853</td>\n",
       "      <td>0.641088</td>\n",
       "      <td>0.844000</td>\n",
       "      <td>0.695280</td>\n",
       "      <td>0.598571</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>70</td>\n",
       "      <td>1.380784</td>\n",
       "      <td>0.610880</td>\n",
       "      <td>0.840200</td>\n",
       "      <td>0.739890</td>\n",
       "      <td>0.640892</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>71</td>\n",
       "      <td>1.233877</td>\n",
       "      <td>0.601307</td>\n",
       "      <td>0.846300</td>\n",
       "      <td>0.679588</td>\n",
       "      <td>0.554288</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>72</td>\n",
       "      <td>1.277399</td>\n",
       "      <td>0.597602</td>\n",
       "      <td>0.851100</td>\n",
       "      <td>0.700541</td>\n",
       "      <td>0.576857</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>73</td>\n",
       "      <td>1.229229</td>\n",
       "      <td>0.593645</td>\n",
       "      <td>0.851200</td>\n",
       "      <td>0.655583</td>\n",
       "      <td>0.573646</td>\n",
       "      <td>03:13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>74</td>\n",
       "      <td>1.262444</td>\n",
       "      <td>0.581748</td>\n",
       "      <td>0.852500</td>\n",
       "      <td>0.694619</td>\n",
       "      <td>0.567825</td>\n",
       "      <td>03:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>75</td>\n",
       "      <td>1.217210</td>\n",
       "      <td>0.624349</td>\n",
       "      <td>0.845600</td>\n",
       "      <td>0.658047</td>\n",
       "      <td>0.559162</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>76</td>\n",
       "      <td>1.216304</td>\n",
       "      <td>0.596451</td>\n",
       "      <td>0.852600</td>\n",
       "      <td>0.654573</td>\n",
       "      <td>0.561730</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>77</td>\n",
       "      <td>1.206554</td>\n",
       "      <td>0.615754</td>\n",
       "      <td>0.848500</td>\n",
       "      <td>0.673398</td>\n",
       "      <td>0.533154</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>78</td>\n",
       "      <td>1.124383</td>\n",
       "      <td>0.626952</td>\n",
       "      <td>0.852800</td>\n",
       "      <td>0.635324</td>\n",
       "      <td>0.489058</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>79</td>\n",
       "      <td>1.166015</td>\n",
       "      <td>0.595053</td>\n",
       "      <td>0.855800</td>\n",
       "      <td>0.642959</td>\n",
       "      <td>0.523055</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>80</td>\n",
       "      <td>1.142673</td>\n",
       "      <td>0.603446</td>\n",
       "      <td>0.853500</td>\n",
       "      <td>0.650097</td>\n",
       "      <td>0.492575</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>81</td>\n",
       "      <td>1.203512</td>\n",
       "      <td>0.578155</td>\n",
       "      <td>0.855900</td>\n",
       "      <td>0.651276</td>\n",
       "      <td>0.552235</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>82</td>\n",
       "      <td>1.187233</td>\n",
       "      <td>0.590109</td>\n",
       "      <td>0.855000</td>\n",
       "      <td>0.655186</td>\n",
       "      <td>0.532046</td>\n",
       "      <td>03:13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>83</td>\n",
       "      <td>1.257218</td>\n",
       "      <td>0.567332</td>\n",
       "      <td>0.858500</td>\n",
       "      <td>0.699557</td>\n",
       "      <td>0.557660</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>84</td>\n",
       "      <td>1.172977</td>\n",
       "      <td>0.591229</td>\n",
       "      <td>0.855800</td>\n",
       "      <td>0.660470</td>\n",
       "      <td>0.512506</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>85</td>\n",
       "      <td>1.185862</td>\n",
       "      <td>0.566091</td>\n",
       "      <td>0.859300</td>\n",
       "      <td>0.666820</td>\n",
       "      <td>0.519040</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>86</td>\n",
       "      <td>1.093079</td>\n",
       "      <td>0.621572</td>\n",
       "      <td>0.850500</td>\n",
       "      <td>0.623498</td>\n",
       "      <td>0.469580</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>87</td>\n",
       "      <td>1.215380</td>\n",
       "      <td>0.569828</td>\n",
       "      <td>0.857200</td>\n",
       "      <td>0.670289</td>\n",
       "      <td>0.545089</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>88</td>\n",
       "      <td>1.133478</td>\n",
       "      <td>0.595722</td>\n",
       "      <td>0.857900</td>\n",
       "      <td>0.649603</td>\n",
       "      <td>0.483874</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>89</td>\n",
       "      <td>1.184425</td>\n",
       "      <td>0.603292</td>\n",
       "      <td>0.855800</td>\n",
       "      <td>0.671753</td>\n",
       "      <td>0.512671</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>90</td>\n",
       "      <td>1.047582</td>\n",
       "      <td>0.599773</td>\n",
       "      <td>0.858400</td>\n",
       "      <td>0.598792</td>\n",
       "      <td>0.448789</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>91</td>\n",
       "      <td>1.153911</td>\n",
       "      <td>0.576635</td>\n",
       "      <td>0.860300</td>\n",
       "      <td>0.661952</td>\n",
       "      <td>0.491957</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>92</td>\n",
       "      <td>1.127042</td>\n",
       "      <td>0.594206</td>\n",
       "      <td>0.855800</td>\n",
       "      <td>0.644425</td>\n",
       "      <td>0.482616</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>93</td>\n",
       "      <td>1.117049</td>\n",
       "      <td>0.581767</td>\n",
       "      <td>0.858200</td>\n",
       "      <td>0.625440</td>\n",
       "      <td>0.491607</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>94</td>\n",
       "      <td>1.005259</td>\n",
       "      <td>0.609383</td>\n",
       "      <td>0.858300</td>\n",
       "      <td>0.587461</td>\n",
       "      <td>0.417797</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>95</td>\n",
       "      <td>1.128302</td>\n",
       "      <td>0.606279</td>\n",
       "      <td>0.858500</td>\n",
       "      <td>0.645155</td>\n",
       "      <td>0.483146</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>96</td>\n",
       "      <td>1.091513</td>\n",
       "      <td>0.600031</td>\n",
       "      <td>0.856800</td>\n",
       "      <td>0.623353</td>\n",
       "      <td>0.468159</td>\n",
       "      <td>03:11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>97</td>\n",
       "      <td>1.186532</td>\n",
       "      <td>0.581192</td>\n",
       "      <td>0.857300</td>\n",
       "      <td>0.654225</td>\n",
       "      <td>0.532306</td>\n",
       "      <td>03:13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>98</td>\n",
       "      <td>1.114383</td>\n",
       "      <td>0.596895</td>\n",
       "      <td>0.857800</td>\n",
       "      <td>0.631332</td>\n",
       "      <td>0.483050</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>99</td>\n",
       "      <td>1.090401</td>\n",
       "      <td>0.616082</td>\n",
       "      <td>0.856400</td>\n",
       "      <td>0.617933</td>\n",
       "      <td>0.472467</td>\n",
       "      <td>03:12</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(100)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true,
    "hidden": true
   },
   "source": [
    "### Warning"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "You may sometimes get a warning like the one below. If training continues, don't worry about it.\n",
    "However, your training may also get stopped. It seems to be a problem with Pytorch dataloader when using multiprocessing. If this is your case, you may need to modify the parameter num_workers in mixmatch and set it to 0. This however, will have a heavy impact on performance!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T16:29:19.405818Z",
     "start_time": "2019-10-15T16:23:04.855240Z"
    },
    "hidden": true,
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "labeled dataset     :           500 samples\n",
      "unlabeled dataset   :        49,500 samples\n",
      "total train samples :       296,448 samples\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>train_Lx</th>\n",
       "      <th>train_Lu*λ</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>2.086064</td>\n",
       "      <td>2.832476</td>\n",
       "      <td>0.169700</td>\n",
       "      <td>1.591187</td>\n",
       "      <td>0.494875</td>\n",
       "      <td>03:07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>1.312561</td>\n",
       "      <td>2.682193</td>\n",
       "      <td>0.190800</td>\n",
       "      <td>0.875799</td>\n",
       "      <td>0.436760</td>\n",
       "      <td>03:06</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Exception ignored in: <function _MultiProcessingDataLoaderIter.__del__ at 0x7fcee3de8c20>\n",
      "Traceback (most recent call last):\n",
      "  File \"/home/oguizadl/anaconda3/envs/fastai-v1/lib/python3.7/site-packages/torch/utils/data/dataloader.py\", line 926, in __del__\n",
      "    self._shutdown_workers()\n",
      "  File \"/home/oguizadl/anaconda3/envs/fastai-v1/lib/python3.7/site-packages/torch/utils/data/dataloader.py\", line 906, in _shutdown_workers\n",
      "    w.join()\n",
      "  File \"/home/oguizadl/anaconda3/envs/fastai-v1/lib/python3.7/multiprocessing/process.py\", line 138, in join\n",
      "    assert self._parent_pid == os.getpid(), 'can only join a child process'\n",
      "AssertionError: can only join a child process\n"
     ]
    }
   ],
   "source": [
    "model = models.WideResNet(num_groups=3, N=4, num_classes=10, k=2, start_nf=32)\n",
    "learn = Learner(data, model, metrics=accuracy).mixmatch(ulist, α=.75, λ=75, num_workers=None)\n",
    "learn.fit_one_cycle(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T16:51:51.961362Z",
     "start_time": "2019-10-15T16:51:21.858933Z"
    },
    "hidden": true,
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "model = models.WideResNet(num_groups=3, N=4, num_classes=10, k=2, start_nf=32)\n",
    "learn = Learner(data, model, metrics=accuracy).mixmatch(ulist, α=.75, λ=75, num_workers=0)\n",
    "learn.fit_one_cycle(2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true
   },
   "source": [
    "## Conclusions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "MixMatch is a state-of-the-art semi-supervised learning technique that works extremely well in those cases where you have a relatively small labeled dataset, and a larger unlabeled dataset.\n",
    "In terms of accuracy, here's the comparison between the previous 4 options: \n",
    "\n",
    "1. Supervised training: 50k labeled samples\n",
    "2. Supervised training: 500 labeled samples (1% of total)\n",
    "3. Supervised training: 500 labeled samples with Mixup\n",
    "4. Semi-supervised training: 500 labeled + 49500 unlabeled samples with MixMatch"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "<img src=\"./images/MixMatch_comparison_performance.jpg\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "The training time however is the price you'll need to pay to use this technique. \n",
    "It takes 4.5 times longer to run mixmatch for 50k (500 labeled + 49.5k unlabeled) than 50k labeled. This seems to be in line with the time required by Google's original MixMatch code. The mention in their website it took them 20+ hours to run on a V100. \n",
    "This is the reason I've only run 100 epochs, and didn't try to replicate the results in the original paper. \n",
    "In the original MixMatch code, they run the equivalent to 450 epochs approx. I only have access to a single, preemptible GPU P100. It would have taken me more than 24h to run."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:45:46.907750Z",
     "start_time": "2019-10-15T17:45:46.903365Z"
    },
    "hidden": true
   },
   "source": [
    "<img src=\"./images/MixMatch_time_comparison.jpg\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-15T17:23:38.614396Z",
     "start_time": "2019-10-15T17:22:42.997356Z"
    },
    "hidden": true
   },
   "source": [
    "So you'll need to make a trade off, between time spend and accuracy. \n",
    "\n",
    "But I still think it's amazing you can get this level of performance with **just 1% of the labels!!**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "hidden": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "fastai-v1",
   "language": "python",
   "name": "fastai-v1"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}