jt-kaggle-plant-seedling-classification-1-Copy1.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Image classification  - Kaggle Plant Seedling Classification"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Put these at the top of every notebook, to get automatic reloading and inline plotting\n",
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here we import the libraries we need"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# This file contains all the main external libs we'll use\n",
    "from fastai.imports import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai.transforms import *\n",
    "from fastai.conv_learner import *\n",
    "from fastai.model import *\n",
    "from fastai.dataset import *\n",
    "from fastai.sgdr import *\n",
    "from fastai.plots import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "PATH = \"data/plant-seedlings-classification/\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "torch.cuda.is_available()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "torch.backends.cudnn.enabled"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "1. Enable data augmentation, and precompute=True\n",
    "1. Use `lr_find()` to find highest learning rate where loss is still clearly improving\n",
    "1. Train last layer from precomputed activations for 1-2 epochs\n",
    "1. Train last layer with data augmentation (i.e. precompute=False) for 2-3 epochs with cycle_len=1\n",
    "1. Unfreeze all layers\n",
    "1. Set earlier layers to 3x-10x lower learning rate than next higher layer\n",
    "1. Use `lr_find()` again\n",
    "1. Train full network with cycle_mult=2 until over-fitting"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## The dataset \n",
    "This is provided from kaggle in the plant-seedlings-classfication competition. The .zip files provided are in test and train foldes with the train folder having subfolders for each class (Keras style).\n",
    "\n",
    "Lets first download the data. Im using the official kaggle CLI https://github.com/Kaggle/kaggle-api"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !mkdir $PATH\n",
    "# !kaggle competitions download -c plant-seedlings-classification -p ~/github/fastai/courses/dl1/data/plant-seedlings-classification"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Unzip all files provided by Kaggle in to directory and then remove the .zip files"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !unzip -q '{PATH}*.zip' -d {PATH}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !rm -r {PATH}*.zip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['train', 'tmp', 'models', 'test', 'labels.csv', 'sample_submission.csv']"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "os.listdir(PATH)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Lets create a csv reference file so we can use the much easier `ImageClassifierData.from_csv()` option"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# from glob2 import glob\n",
    "# import pandas as pd\n",
    "\n",
    "# df = pd.DataFrame(columns=[\"file\", \"species\"])\n",
    "\n",
    "# for image in glob(f'{PATH}train/**/*.png'):\n",
    "#     dir_ = image.split('/')\n",
    "#     file_, species = dir_[-1], dir_[-2]\n",
    "\n",
    "#     df = df.append({\n",
    "#         \"file\": file_,\n",
    "#         \"species\": species.replace(\" \", \"_\")\n",
    "#         }, ignore_index=True)\n",
    "\n",
    "# df.to_csv(f'{PATH}labels.csv', index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Read in our dataframe and view the structure"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# df  = pd.read_csv(f'{PATH}labels.csv')\n",
    "\n",
    "# df['species'] = df['species'].replace(\" \", \"_\", regex=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# df.to_csv(f'{PATH}labels.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "labels_csv = f'{PATH}labels.csv'\n",
    "labels_df = pd.read_csv(labels_csv)\n",
    "\n",
    "# labels_df.head(10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now move all the images from under the class subfolders into the root train folder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !mv {PATH}train/**/*.png {PATH}train/"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Remove all the empty class subfolders from train/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !rm -r {PATH}train/**/"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Lets view the distribution of the classes a a pivot`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>file</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>species</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Loose_Silky-bent</th>\n",
       "      <td>654</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Common_Chickweed</th>\n",
       "      <td>611</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Scentless_Mayweed</th>\n",
       "      <td>516</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Small-flowered_Cranesbill</th>\n",
       "      <td>496</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fat_Hen</th>\n",
       "      <td>475</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Charlock</th>\n",
       "      <td>390</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Sugar_beet</th>\n",
       "      <td>385</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Cleavers</th>\n",
       "      <td>287</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Black-grass</th>\n",
       "      <td>263</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Shepherds_Purse</th>\n",
       "      <td>231</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Common_wheat</th>\n",
       "      <td>221</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Maize</th>\n",
       "      <td>221</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                           file\n",
       "species                        \n",
       "Loose_Silky-bent            654\n",
       "Common_Chickweed            611\n",
       "Scentless_Mayweed           516\n",
       "Small-flowered_Cranesbill   496\n",
       "Fat_Hen                     475\n",
       "Charlock                    390\n",
       "Sugar_beet                  385\n",
       "Cleavers                    287\n",
       "Black-grass                 263\n",
       "Shepherds_Purse             231\n",
       "Common_wheat                221\n",
       "Maize                       221"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "labels_df.pivot_table(index = 'species', aggfunc = len).sort_values('file', ascending = False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Take a look at the dataset distribution size of images. we first need to define the data loader and transformations. We can later change these or wrap in a data function before training begins\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Initial Model parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "arch=resnext50 # model architecture to use\n",
    "sz=224 # image size\n",
    "bs=32 # mini-batch size"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Make validation indexes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create validation indexes\n",
    "n = len(list(open(labels_csv)))-1\n",
    "val_idxs = get_cv_idxs(n)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Transformations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Data Augementation/Transformation Settings\n",
    "tfms = tfms_from_model(arch, # pass in architecture\n",
    "                       sz, # image size\n",
    "                       aug_tfms=transforms_top_down, # side_on for normal images top_down for satelite of photos taken from above\n",
    "                       max_zoom=1.15)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Data Generator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = ImageClassifierData.from_csv(path = PATH,\n",
    "                                    folder='train',\n",
    "                                    csv_fname = labels_csv, # name of the labels reference file\n",
    "                                    val_idxs = val_idxs, # validation indexes\n",
    "                                    test_name = 'test',\n",
    "                                    tfms = tfms, # transformations to apply\n",
    "                                    bs = bs) #mini batch size"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Data set image resolutions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([2604.,  897.,  215.,   59.,   13.,    4.,    3.,    3.,    0.,    2.]),\n",
       " array([  49. ,  443.2,  837.4, 1231.6, 1625.8, 2020. , 2414.2, 2808.4, 3202.6, 3596.8, 3991. ]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# see distribution of all images\n",
    "size_d = {k: Image.open(PATH + k).size for k in data.trn_ds.fnames}\n",
    "row_sz, col_sz = list(zip(*size_d.values()))\n",
    "row_sz = np.array(row_sz); col_sz = np.array(col_sz)\n",
    "plt.hist(row_sz)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([1008.,  832.,  269.,  431.,  371.,  261.,  193.,  109.,   80.,   64.]),\n",
       " array([ 49., 144., 239., 334., 429., 524., 619., 714., 809., 904., 999.]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# see distribution of images less than 1000 pixels\n",
    "plt.hist(row_sz[row_sz < 1000])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true
   },
   "source": [
    "## Model Build"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Uncomment the below if you need to reset your precomputed activations\n",
    "# shutil.rmtree(f'{PATH}tmp', ignore_errors=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# shutil.rmtree(f'{PATH}model', ignore_errors=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "hidden": true,
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "# Create Learner\n",
    "learn = ConvLearner.pretrained(arch,\n",
    "                               data,\n",
    "                               precompute=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Find initial learning rate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "eeb26c92a57c4a21a60ad0ffdeb398a0",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 78%|███████▊  | 47/60 [00:01<00:00, 40.14it/s, loss=26.3]\n",
      "                                                          \r"
     ]
    }
   ],
   "source": [
    "# use learn.lr_find to find the optimal learning rate\n",
    "lrf = learn.lr_find()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#plot learing rate\n",
    "learn.sched.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr = 1e-1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "74cb476a16124fa09821b63c85a0ae93",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                 \n",
      "    0      1.471899   0.824452   0.747368  \n",
      "    1      0.975393   0.532872   0.821053                  \n",
      "    2      0.683023   0.439332   0.847368                  \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([0.43933]), 0.8473684205506977]"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Fit learner with a few epochs\n",
    "learn.fit(lr, 3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.precompute=False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "fed6a1e458b54a44b64cfd6ccba0de33",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=3), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                  \n",
      "    0      0.633013   0.379534   0.871579  \n",
      "    1      0.586553   0.359982   0.878947                  \n",
      "    2      0.54863    0.353986   0.877895                  \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([0.35399]), 0.8778947384733903]"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lr, 3, cycle_len=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('224_lastlayer_resnext50') # training and validation sets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('224_lastlayer_resnext50') # training and validation sets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.unfreeze()\n",
    "learn.bn_freeze(True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "2b7a0779efdc4c6788fa09ed7f3faff2",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=1), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 34%|███▎      | 40/119 [00:45<01:30,  1.14s/it, loss=2.15] "
     ]
    }
   ],
   "source": [
    "lrf = learn.lr_find()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#plot learing rate\n",
    "learn.sched.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "lrs=np.array([1e-6,1e-4,1e-3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8e856dd92379464bb940035ce3d4f59f",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=7), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                      \n",
      "    0      0.338084   0.25202    0.910526  \n",
      "    1      0.337858   0.221417   0.92                        \n",
      "    2      0.205734   0.19404    0.929474                    \n",
      "    3      0.222315   0.192694   0.929474                    \n",
      "    4      0.187601   0.185757   0.934737                    \n",
      "    5      0.157373   0.162369   0.942105                    \n",
      "    6      0.130476   0.153369   0.947368                    \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([0.15337]), 0.9473684206761812]"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lrs, 3, cycle_len=1, cycle_mult=2, cycle_save_name=\"plantseed_resnext\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('224_all_lr432_1')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.load('224_all_lr432_1')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "fad8d54102f84cb6976c9a005fa88b75",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, description='Epoch', max=2), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch      trn_loss   val_loss   accuracy                    \n",
      "    0      0.149768   0.143184   0.949474  \n",
      "    1      0.144236   0.154117   0.943158                    \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[array([0.15412]), 0.9431578943603918]"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.fit(lrs, 2, cycle_len=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                             \r"
     ]
    }
   ],
   "source": [
    "log_preds,y = learn.TTA(True)\n",
    "probs = np.mean(np.exp(log_preds),0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9610526315789474"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# validation accuracy after TTA\n",
    "accuracy_np(probs, y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds = np.argmax(probs, axis=1)\n",
    "probs = probs[:,1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "hidden": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "cm = confusion_matrix(y, preds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[ 33   0   0   0   0   0  12   0   1   0   0   0]\n",
      " [  0  89   0   0   1   0   0   0   0   0   0   0]\n",
      " [  0   0  46   0   0   0   0   0   0   0   0   0]\n",
      " [  0   0   0 122   0   0   0   0   0   0   0   0]\n",
      " [  1   0   0   0  39   1   0   1   0   0   0   0]\n",
      " [  0   0   0   0   0  93   0   0   0   0   0   0]\n",
      " [ 15   0   0   0   0   0 116   0   0   0   0   0]\n",
      " [  0   0   0   0   0   0   0  53   0   0   0   0]\n",
      " [  0   0   0   0   0   0   0   0  93   0   0   0]\n",
      " [  0   0   0   0   0   0   0   0   3  45   0   0]\n",
      " [  0   0   0   0   0   0   1   0   0   0 102   0]\n",
      " [  0   0   0   1   0   0   0   0   0   0   0  82]]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_confusion_matrix(cm, data.classes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Analyzing results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true
   },
   "source": [
    "### Confusion matrix "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "hidden": true
   },
   "source": [
    "We can just print out the confusion matrix, or we can show a graphical view (which is mainly useful for dependents with a larger number of categories)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Predict the test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                                             \r"
     ]
    }
   ],
   "source": [
    "log_preds, y = learn.TTA(is_test=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds = np.mean(np.exp(log_preds),0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(794, 12)"
      ]
     },
     "execution_count": 85,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds = np.argmax(preds, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(794,)"
      ]
     },
     "execution_count": 87,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds_classes = [data.classes[i].replace(\"_\", \" \") for i in preds]\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [],
   "source": [
    "submission = pd.DataFrame(preds_classes, os.listdir(f'{PATH}test'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {},
   "outputs": [],
   "source": [
    "submission.to_csv(f'{PATH}tmp/subm.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<a href='data/plant-seedlings-classification/tmp/subm.csv' target='_blank'>data/plant-seedlings-classification/tmp/subm.csv</a><br>"
      ],
      "text/plain": [
       "/home/ubuntu/github/fastai/courses/dl1/data/plant-seedlings-classification/tmp/subm.csv"
      ]
     },
     "execution_count": 110,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from IPython.display import FileLink\n",
    "FileLink(f'{PATH}tmp/subm.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(950, 1)"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "ename": "ValueError",
     "evalue": "Length mismatch: Expected axis has 1 elements, new values have 12 elements",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mValueError\u001b[0m                                Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-53-fc1b59b5d0db>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0mds\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mpd\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mDataFrame\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mprobs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0mds\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcolumns\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mdata\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mclasses\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m~/anaconda3/envs/fastai/lib/python3.6/site-packages/pandas/core/generic.py\u001b[0m in \u001b[0;36m__setattr__\u001b[0;34m(self, name, value)\u001b[0m\n\u001b[1;32m   3625\u001b[0m         \u001b[0;32mtry\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3626\u001b[0m             \u001b[0mobject\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m__getattribute__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3627\u001b[0;31m             \u001b[0;32mreturn\u001b[0m \u001b[0mobject\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m__setattr__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mname\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3628\u001b[0m         \u001b[0;32mexcept\u001b[0m \u001b[0mAttributeError\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3629\u001b[0m             \u001b[0;32mpass\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32mpandas/_libs/properties.pyx\u001b[0m in \u001b[0;36mpandas._libs.properties.AxisProperty.__set__\u001b[0;34m()\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/envs/fastai/lib/python3.6/site-packages/pandas/core/generic.py\u001b[0m in \u001b[0;36m_set_axis\u001b[0;34m(self, axis, labels)\u001b[0m\n\u001b[1;32m    557\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    558\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m_set_axis\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlabels\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 559\u001b[0;31m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_data\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mset_axis\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0maxis\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlabels\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    560\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_clear_item_cache\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    561\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/envs/fastai/lib/python3.6/site-packages/pandas/core/internals.py\u001b[0m in \u001b[0;36mset_axis\u001b[0;34m(self, axis, new_labels)\u001b[0m\n\u001b[1;32m   3072\u001b[0m             raise ValueError('Length mismatch: Expected axis has %d elements, '\n\u001b[1;32m   3073\u001b[0m                              \u001b[0;34m'new values have %d elements'\u001b[0m \u001b[0;34m%\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3074\u001b[0;31m                              (old_len, new_len))\n\u001b[0m\u001b[1;32m   3075\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3076\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0maxes\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0maxis\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnew_labels\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mValueError\u001b[0m: Length mismatch: Expected axis has 1 elements, new values have 12 elements"
     ]
    }
   ],
   "source": [
    "ds = pd.DataFrame(probs)\n",
    "ds.columns = data.classes"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Prepare submission"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First we need to have a look at the sample submission file to see what layout the file needs to be in"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "samp_sub_df = pd.read_csv(f'{PATH}sample_submission.csv')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "From above we can see that we need a filename andthe predicted class `species`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>file</th>\n",
       "      <th>species</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0021e90e4.png</td>\n",
       "      <td>Sugar beet</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>003d61042.png</td>\n",
       "      <td>Sugar beet</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>007b3da8b.png</td>\n",
       "      <td>Sugar beet</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0086a6340.png</td>\n",
       "      <td>Sugar beet</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>00c47e980.png</td>\n",
       "      <td>Sugar beet</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            file     species\n",
       "0  0021e90e4.png  Sugar beet\n",
       "1  003d61042.png  Sugar beet\n",
       "2  007b3da8b.png  Sugar beet\n",
       "3  0086a6340.png  Sugar beet\n",
       "4  00c47e980.png  Sugar beet"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "samp_sub_df.head()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:fastai]",
   "language": "python",
   "name": "conda-env-fastai-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  },
  "nbpresent": {
   "slides": {},
   "themes": {
    "default": "25df69a3-8a9e-40c3-bec1-7d78b14fe19a",
    "theme": {
     "25df69a3-8a9e-40c3-bec1-7d78b14fe19a": {
      "backgrounds": {
       "dc7afa04-bf90-40b1-82a5-726e3cff5267": {
        "background-color": "31af15d2-7e15-44c5-ab5e-e04b16a89eff",
        "id": "dc7afa04-bf90-40b1-82a5-726e3cff5267"
       }
      },
      "id": "25df69a3-8a9e-40c3-bec1-7d78b14fe19a",
      "palette": {
       "19cc588f-0593-49c9-9f4b-e4d7cc113b1c": {
        "id": "19cc588f-0593-49c9-9f4b-e4d7cc113b1c",
        "rgb": [
         252,
         252,
         252
        ]
       },
       "31af15d2-7e15-44c5-ab5e-e04b16a89eff": {
        "id": "31af15d2-7e15-44c5-ab5e-e04b16a89eff",
        "rgb": [
         68,
         68,
         68
        ]
       },
       "50f92c45-a630-455b-aec3-788680ec7410": {
        "id": "50f92c45-a630-455b-aec3-788680ec7410",
        "rgb": [
         197,
         226,
         245
        ]
       },
       "c5cc3653-2ee1-402a-aba2-7caae1da4f6c": {
        "id": "c5cc3653-2ee1-402a-aba2-7caae1da4f6c",
        "rgb": [
         43,
         126,
         184
        ]
       },
       "efa7f048-9acb-414c-8b04-a26811511a21": {
        "id": "efa7f048-9acb-414c-8b04-a26811511a21",
        "rgb": [
         25.118061674008803,
         73.60176211453744,
         107.4819383259912
        ]
       }
      },
      "rules": {
       "a": {
        "color": "19cc588f-0593-49c9-9f4b-e4d7cc113b1c"
       },
       "blockquote": {
        "color": "50f92c45-a630-455b-aec3-788680ec7410",
        "font-size": 3
       },
       "code": {
        "font-family": "Anonymous Pro"
       },
       "h1": {
        "color": "19cc588f-0593-49c9-9f4b-e4d7cc113b1c",
        "font-family": "Merriweather",
        "font-size": 8
       },
       "h2": {
        "color": "19cc588f-0593-49c9-9f4b-e4d7cc113b1c",
        "font-family": "Merriweather",
        "font-size": 6
       },
       "h3": {
        "color": "50f92c45-a630-455b-aec3-788680ec7410",
        "font-family": "Lato",
        "font-size": 5.5
       },
       "h4": {
        "color": "c5cc3653-2ee1-402a-aba2-7caae1da4f6c",
        "font-family": "Lato",
        "font-size": 5
       },
       "h5": {
        "font-family": "Lato"
       },
       "h6": {
        "font-family": "Lato"
       },
       "h7": {
        "font-family": "Lato"
       },
       "li": {
        "color": "50f92c45-a630-455b-aec3-788680ec7410",
        "font-size": 3.25
       },
       "pre": {
        "font-family": "Anonymous Pro",
        "font-size": 4
       }
      },
      "text-base": {
       "color": "19cc588f-0593-49c9-9f4b-e4d7cc113b1c",
       "font-family": "Lato",
       "font-size": 4
      }
     }
    }
   }
  },
  "toc": {
   "colors": {
    "hover_highlight": "#DAA520",
    "navigate_num": "#000000",
    "navigate_text": "#333333",
    "running_highlight": "#FF0000",
    "selected_highlight": "#FFD700",
    "sidebar_border": "#EEEEEE",
    "wrapper_background": "#FFFFFF"
   },
   "moveMenuLeft": true,
   "nav_menu": {
    "height": "266px",
    "width": "252px"
   },
   "navigate_menu": true,
   "number_sections": true,
   "sideBar": true,
   "threshold": 4,
   "toc_cell": false,
   "toc_section_display": "block",
   "toc_window_display": false,
   "widenNotebook": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}