sound_classification.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "#zhangjianming audio classification via 3 features 2019-12-16\n",
    "\n",
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline\n",
    "import os\n",
    "os.environ['CUDA_VISIBLE_DEVICES']='0'\n",
    "from fastai.vision import *\n",
    "from fastai.vision.models.wrn import wrn_22\n",
    "from fastai.vision.models import *\n",
    "\n",
    "from torchvision.models import *\n",
    "import pretrainedmodels\n",
    "from efficientnet_pytorch import EfficientNet\n",
    "SEED = 2019\n",
    "def seed_everything(seed):\n",
    "    random.seed(seed)\n",
    "    os.environ['PYTHONHASHSEED'] = str(seed)\n",
    "    np.random.seed(seed)\n",
    "    torch.manual_seed(seed)\n",
    "    torch.cuda.manual_seed(seed)\n",
    "    torch.backends.cudnn.deterministic = True\n",
    "    torch.backends.cudnn.benchmark = True\n",
    "\n",
    "seed_everything(SEED)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "path = datapath4file('/data/zhangjianming/kaggle/sound/data/melspec/')\n",
    "#path = datapath4file('/data/zhangjianming/kaggle/sound/data/mfccs/')\n",
    "#path = datapath4file('/data/zhangjianming/kaggle/sound/data/chroma/')\n",
    "\n",
    "3 features \n",
    "\n",
    "def get_data(sz,bs):\n",
    "    tfms = get_transforms(flip_vert=False,max_rotate=10, p_affine =1., p_lighting=1.,\n",
    "                      max_lighting=0.7, max_zoom=1.5, max_warp=0.1) \n",
    "    #tfms = get_transforms()   \n",
    "    data = ImageDataBunch.from_folder(path, size=sz,train = 'train', valid_pct = 0.2, test = 'test', bs = bs)\n",
    "    data.normalize(imagenet_stats)\n",
    "    return data\n",
    "path.ls()\n",
    "data = get_data(224,32)\n",
    "#data.show_batch(row =6,figsize = (15,15))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loaded pretrained weights for efficientnet-b0\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "Learner(data=ImageDataBunch;\n",
       "\n",
       "Train: LabelList (1965 items)\n",
       "x: ImageList\n",
       "Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112)\n",
       "y: CategoryList\n",
       "32,32,32,32,32\n",
       "Path: /data/zhangjianming/kaggle/sound/data/train_data_new;\n",
       "\n",
       "Valid: LabelList (1965 items)\n",
       "x: ImageList\n",
       "Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112)\n",
       "y: CategoryList\n",
       "32,32,32,32,32\n",
       "Path: /data/zhangjianming/kaggle/sound/data/train_data_new;\n",
       "\n",
       "Test: LabelList (400 items)\n",
       "x: ImageList\n",
       "Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112)\n",
       "y: EmptyLabelList\n",
       ",,,,\n",
       "Path: /data/zhangjianming/kaggle/sound/data/train_data_new, model=EfficientNet(\n",
       "  (_conv_stem): Conv2dStaticSamePadding(\n",
       "    3, 32, kernel_size=(3, 3), stride=(2, 2), bias=False\n",
       "    (static_padding): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "  )\n",
       "  (_bn0): BatchNorm2d(32, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (_blocks): ModuleList(\n",
       "    (0): MBConvBlock(\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        32, 32, kernel_size=(3, 3), stride=[1, 1], groups=32, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(32, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        32, 8, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        8, 32, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        32, 16, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(16, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (1): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        16, 96, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(96, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        96, 96, kernel_size=(3, 3), stride=[2, 2], groups=96, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(96, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        96, 4, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        4, 96, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        96, 24, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(24, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (2): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        24, 144, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        144, 144, kernel_size=(3, 3), stride=(1, 1), groups=144, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        144, 6, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        6, 144, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        144, 24, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(24, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (3): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        24, 144, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        144, 144, kernel_size=(5, 5), stride=[2, 2], groups=144, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 2, 1, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        144, 6, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        6, 144, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        144, 40, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(40, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (4): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        240, 240, kernel_size=(5, 5), stride=(1, 1), groups=240, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        240, 10, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        10, 240, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        240, 40, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(40, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (5): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        240, 240, kernel_size=(3, 3), stride=[2, 2], groups=240, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        240, 10, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        10, 240, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        240, 80, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (6): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        480, 480, kernel_size=(3, 3), stride=(1, 1), groups=480, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        480, 20, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        20, 480, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        480, 80, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (7): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        480, 480, kernel_size=(3, 3), stride=(1, 1), groups=480, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        480, 20, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        20, 480, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        480, 80, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (8): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        480, 480, kernel_size=(5, 5), stride=[1, 1], groups=480, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        480, 20, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        20, 480, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        480, 112, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (9): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        672, 672, kernel_size=(5, 5), stride=(1, 1), groups=672, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        672, 28, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        28, 672, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        672, 112, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (10): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        672, 672, kernel_size=(5, 5), stride=(1, 1), groups=672, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        672, 28, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        28, 672, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        672, 112, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (11): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        672, 672, kernel_size=(5, 5), stride=[2, 2], groups=672, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 2, 1, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        672, 28, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        28, 672, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        672, 192, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (12): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        1152, 1152, kernel_size=(5, 5), stride=(1, 1), groups=1152, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        1152, 48, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        48, 1152, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (13): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        1152, 1152, kernel_size=(5, 5), stride=(1, 1), groups=1152, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        1152, 48, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        48, 1152, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (14): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        1152, 1152, kernel_size=(5, 5), stride=(1, 1), groups=1152, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        1152, 48, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        48, 1152, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (15): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        1152, 1152, kernel_size=(3, 3), stride=[1, 1], groups=1152, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        1152, 48, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        48, 1152, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        1152, 320, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(320, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "  )\n",
       "  (_conv_head): Conv2dStaticSamePadding(\n",
       "    320, 1280, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "    (static_padding): Identity()\n",
       "  )\n",
       "  (_bn1): BatchNorm2d(1280, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (_avg_pooling): AdaptiveAvgPool2d(output_size=1)\n",
       "  (_dropout): Dropout(p=0.2, inplace=False)\n",
       "  (_fc): Linear(in_features=1280, out_features=50, bias=True)\n",
       "  (_swish): MemoryEfficientSwish()\n",
       "), opt_func=functools.partial(<class 'torch.optim.adam.Adam'>, betas=(0.9, 0.99)), loss_func=FlattenedLoss of CrossEntropyLoss(), metrics=[<function accuracy at 0x7feaf60da8c8>], true_wd=True, bn_wd=True, wd=0.01, train_bn=True, path=PosixPath('/data/zhangjianming/kaggle/sound/data/train_data_new'), model_dir='models', callback_fns=[functools.partial(<class 'fastai.basic_train.Recorder'>, add_time=True, silent=False), functools.partial(<class 'fastai.callbacks.mixup.MixUpCallback'>, alpha=0.4, stack_x=False, stack_y=True)], callbacks=[MixedPrecision\n",
       "learn: Learner(data=ImageDataBunch;\n",
       "\n",
       "Train: LabelList (1965 items)\n",
       "x: ImageList\n",
       "Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112)\n",
       "y: CategoryList\n",
       "32,32,32,32,32\n",
       "Path: /data/zhangjianming/kaggle/sound/data/train_data_new;\n",
       "\n",
       "Valid: LabelList (1965 items)\n",
       "x: ImageList\n",
       "Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112)\n",
       "y: CategoryList\n",
       "32,32,32,32,32\n",
       "Path: /data/zhangjianming/kaggle/sound/data/train_data_new;\n",
       "\n",
       "Test: LabelList (400 items)\n",
       "x: ImageList\n",
       "Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112),Image (3, 112, 112)\n",
       "y: EmptyLabelList\n",
       ",,,,\n",
       "Path: /data/zhangjianming/kaggle/sound/data/train_data_new, model=EfficientNet(\n",
       "  (_conv_stem): Conv2dStaticSamePadding(\n",
       "    3, 32, kernel_size=(3, 3), stride=(2, 2), bias=False\n",
       "    (static_padding): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "  )\n",
       "  (_bn0): BatchNorm2d(32, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (_blocks): ModuleList(\n",
       "    (0): MBConvBlock(\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        32, 32, kernel_size=(3, 3), stride=[1, 1], groups=32, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(32, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        32, 8, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        8, 32, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        32, 16, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(16, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (1): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        16, 96, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(96, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        96, 96, kernel_size=(3, 3), stride=[2, 2], groups=96, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(96, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        96, 4, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        4, 96, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        96, 24, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(24, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (2): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        24, 144, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        144, 144, kernel_size=(3, 3), stride=(1, 1), groups=144, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        144, 6, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        6, 144, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        144, 24, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(24, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (3): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        24, 144, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        144, 144, kernel_size=(5, 5), stride=[2, 2], groups=144, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 2, 1, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        144, 6, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        6, 144, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        144, 40, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(40, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (4): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        240, 240, kernel_size=(5, 5), stride=(1, 1), groups=240, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        240, 10, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        10, 240, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        240, 40, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(40, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (5): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        240, 240, kernel_size=(3, 3), stride=[2, 2], groups=240, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        240, 10, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        10, 240, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        240, 80, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (6): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        480, 480, kernel_size=(3, 3), stride=(1, 1), groups=480, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        480, 20, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        20, 480, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        480, 80, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (7): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        480, 480, kernel_size=(3, 3), stride=(1, 1), groups=480, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        480, 20, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        20, 480, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        480, 80, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (8): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        480, 480, kernel_size=(5, 5), stride=[1, 1], groups=480, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        480, 20, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        20, 480, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        480, 112, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (9): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        672, 672, kernel_size=(5, 5), stride=(1, 1), groups=672, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        672, 28, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        28, 672, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        672, 112, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (10): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        672, 672, kernel_size=(5, 5), stride=(1, 1), groups=672, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        672, 28, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        28, 672, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        672, 112, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (11): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        672, 672, kernel_size=(5, 5), stride=[2, 2], groups=672, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 2, 1, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        672, 28, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        28, 672, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        672, 192, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (12): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        1152, 1152, kernel_size=(5, 5), stride=(1, 1), groups=1152, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        1152, 48, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        48, 1152, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (13): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        1152, 1152, kernel_size=(5, 5), stride=(1, 1), groups=1152, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        1152, 48, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        48, 1152, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (14): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        1152, 1152, kernel_size=(5, 5), stride=(1, 1), groups=1152, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        1152, 48, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        48, 1152, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "    (15): MBConvBlock(\n",
       "      (_expand_conv): Conv2dStaticSamePadding(\n",
       "        192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn0): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_depthwise_conv): Conv2dStaticSamePadding(\n",
       "        1152, 1152, kernel_size=(3, 3), stride=[1, 1], groups=1152, bias=False\n",
       "        (static_padding): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "      )\n",
       "      (_bn1): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_se_reduce): Conv2dStaticSamePadding(\n",
       "        1152, 48, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_se_expand): Conv2dStaticSamePadding(\n",
       "        48, 1152, kernel_size=(1, 1), stride=(1, 1)\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_project_conv): Conv2dStaticSamePadding(\n",
       "        1152, 320, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "        (static_padding): Identity()\n",
       "      )\n",
       "      (_bn2): BatchNorm2d(320, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "      (_swish): MemoryEfficientSwish()\n",
       "    )\n",
       "  )\n",
       "  (_conv_head): Conv2dStaticSamePadding(\n",
       "    320, 1280, kernel_size=(1, 1), stride=(1, 1), bias=False\n",
       "    (static_padding): Identity()\n",
       "  )\n",
       "  (_bn1): BatchNorm2d(1280, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (_avg_pooling): AdaptiveAvgPool2d(output_size=1)\n",
       "  (_dropout): Dropout(p=0.2, inplace=False)\n",
       "  (_fc): Linear(in_features=1280, out_features=50, bias=True)\n",
       "  (_swish): MemoryEfficientSwish()\n",
       "), opt_func=functools.partial(<class 'torch.optim.adam.Adam'>, betas=(0.9, 0.99)), loss_func=FlattenedLoss of CrossEntropyLoss(), metrics=[<function accuracy at 0x7feaf60da8c8>], true_wd=True, bn_wd=True, wd=0.01, train_bn=True, path=PosixPath('/data/zhangjianming/kaggle/sound/data/train_data_new'), model_dir='models', callback_fns=[functools.partial(<class 'fastai.basic_train.Recorder'>, add_time=True, silent=False), functools.partial(<class 'fastai.callbacks.mixup.MixUpCallback'>, alpha=0.4, stack_x=False, stack_y=True)], callbacks=[...], layer_groups=[Sequential(\n",
       "  (0): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "  (1): ParameterModule()\n",
       "  (2): BatchNorm2d(32, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "), Sequential(\n",
       "  (0): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "  (1): ParameterModule()\n",
       "  (2): BatchNorm2d(32, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (3): Identity()\n",
       "  (4): ParameterModule()\n",
       "  (5): ParameterModule()\n",
       "  (6): Identity()\n",
       "  (7): ParameterModule()\n",
       "  (8): ParameterModule()\n",
       "  (9): Identity()\n",
       "  (10): ParameterModule()\n",
       "  (11): BatchNorm2d(16, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (12): MemoryEfficientSwish()\n",
       "  (13): Identity()\n",
       "  (14): ParameterModule()\n",
       "  (15): BatchNorm2d(96, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (16): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "  (17): ParameterModule()\n",
       "  (18): BatchNorm2d(96, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (19): Identity()\n",
       "  (20): ParameterModule()\n",
       "  (21): ParameterModule()\n",
       "  (22): Identity()\n",
       "  (23): ParameterModule()\n",
       "  (24): ParameterModule()\n",
       "  (25): Identity()\n",
       "  (26): ParameterModule()\n",
       "  (27): BatchNorm2d(24, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (28): MemoryEfficientSwish()\n",
       "  (29): Identity()\n",
       "  (30): ParameterModule()\n",
       "  (31): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (32): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "  (33): ParameterModule()\n",
       "  (34): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (35): Identity()\n",
       "  (36): ParameterModule()\n",
       "  (37): ParameterModule()\n",
       "  (38): Identity()\n",
       "  (39): ParameterModule()\n",
       "  (40): ParameterModule()\n",
       "  (41): Identity()\n",
       "  (42): ParameterModule()\n",
       "  (43): BatchNorm2d(24, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (44): MemoryEfficientSwish()\n",
       "  (45): Identity()\n",
       "  (46): ParameterModule()\n",
       "  (47): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (48): ZeroPad2d(padding=(1, 2, 1, 2), value=0.0)\n",
       "  (49): ParameterModule()\n",
       "  (50): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (51): Identity()\n",
       "  (52): ParameterModule()\n",
       "  (53): ParameterModule()\n",
       "  (54): Identity()\n",
       "  (55): ParameterModule()\n",
       "  (56): ParameterModule()\n",
       "  (57): Identity()\n",
       "  (58): ParameterModule()\n",
       "  (59): BatchNorm2d(40, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (60): MemoryEfficientSwish()\n",
       "  (61): Identity()\n",
       "  (62): ParameterModule()\n",
       "  (63): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (64): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (65): ParameterModule()\n",
       "  (66): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (67): Identity()\n",
       "  (68): ParameterModule()\n",
       "  (69): ParameterModule()\n",
       "  (70): Identity()\n",
       "  (71): ParameterModule()\n",
       "  (72): ParameterModule()\n",
       "  (73): Identity()\n",
       "  (74): ParameterModule()\n",
       "  (75): BatchNorm2d(40, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (76): MemoryEfficientSwish()\n",
       "  (77): Identity()\n",
       "  (78): ParameterModule()\n",
       "  (79): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (80): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "  (81): ParameterModule()\n",
       "  (82): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (83): Identity()\n",
       "  (84): ParameterModule()\n",
       "  (85): ParameterModule()\n",
       "  (86): Identity()\n",
       "  (87): ParameterModule()\n",
       "  (88): ParameterModule()\n",
       "  (89): Identity()\n",
       "  (90): ParameterModule()\n",
       "  (91): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (92): MemoryEfficientSwish()\n",
       "  (93): Identity()\n",
       "  (94): ParameterModule()\n",
       "  (95): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (96): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "  (97): ParameterModule()\n",
       "  (98): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (99): Identity()\n",
       "  (100): ParameterModule()\n",
       "  (101): ParameterModule()\n",
       "  (102): Identity()\n",
       "  (103): ParameterModule()\n",
       "  (104): ParameterModule()\n",
       "  (105): Identity()\n",
       "  (106): ParameterModule()\n",
       "  (107): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (108): MemoryEfficientSwish()\n",
       "  (109): Identity()\n",
       "  (110): ParameterModule()\n",
       "  (111): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (112): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "  (113): ParameterModule()\n",
       "  (114): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (115): Identity()\n",
       "  (116): ParameterModule()\n",
       "  (117): ParameterModule()\n",
       "  (118): Identity()\n",
       "  (119): ParameterModule()\n",
       "  (120): ParameterModule()\n",
       "  (121): Identity()\n",
       "  (122): ParameterModule()\n",
       "  (123): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (124): MemoryEfficientSwish()\n",
       "  (125): Identity()\n",
       "  (126): ParameterModule()\n",
       "  (127): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (128): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (129): ParameterModule()\n",
       "  (130): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (131): Identity()\n",
       "  (132): ParameterModule()\n",
       "  (133): ParameterModule()\n",
       "  (134): Identity()\n",
       "  (135): ParameterModule()\n",
       "  (136): ParameterModule()\n",
       "  (137): Identity()\n",
       "  (138): ParameterModule()\n",
       "  (139): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (140): MemoryEfficientSwish()\n",
       "  (141): Identity()\n",
       "  (142): ParameterModule()\n",
       "  (143): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (144): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (145): ParameterModule()\n",
       "  (146): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (147): Identity()\n",
       "  (148): ParameterModule()\n",
       "  (149): ParameterModule()\n",
       "  (150): Identity()\n",
       "  (151): ParameterModule()\n",
       "  (152): ParameterModule()\n",
       "  (153): Identity()\n",
       "  (154): ParameterModule()\n",
       "  (155): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (156): MemoryEfficientSwish()\n",
       "  (157): Identity()\n",
       "  (158): ParameterModule()\n",
       "  (159): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (160): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (161): ParameterModule()\n",
       "  (162): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (163): Identity()\n",
       "  (164): ParameterModule()\n",
       "  (165): ParameterModule()\n",
       "  (166): Identity()\n",
       "  (167): ParameterModule()\n",
       "  (168): ParameterModule()\n",
       "  (169): Identity()\n",
       "  (170): ParameterModule()\n",
       "  (171): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (172): MemoryEfficientSwish()\n",
       "  (173): Identity()\n",
       "  (174): ParameterModule()\n",
       "  (175): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (176): ZeroPad2d(padding=(1, 2, 1, 2), value=0.0)\n",
       "  (177): ParameterModule()\n",
       "  (178): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (179): Identity()\n",
       "  (180): ParameterModule()\n",
       "  (181): ParameterModule()\n",
       "  (182): Identity()\n",
       "  (183): ParameterModule()\n",
       "  (184): ParameterModule()\n",
       "  (185): Identity()\n",
       "  (186): ParameterModule()\n",
       "  (187): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (188): MemoryEfficientSwish()\n",
       "  (189): Identity()\n",
       "  (190): ParameterModule()\n",
       "  (191): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (192): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (193): ParameterModule()\n",
       "  (194): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (195): Identity()\n",
       "  (196): ParameterModule()\n",
       "  (197): ParameterModule()\n",
       "  (198): Identity()\n",
       "  (199): ParameterModule()\n",
       "  (200): ParameterModule()\n",
       "  (201): Identity()\n",
       "  (202): ParameterModule()\n",
       "  (203): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (204): MemoryEfficientSwish()\n",
       "  (205): Identity()\n",
       "  (206): ParameterModule()\n",
       "  (207): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (208): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (209): ParameterModule()\n",
       "  (210): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (211): Identity()\n",
       "  (212): ParameterModule()\n",
       "  (213): ParameterModule()\n",
       "  (214): Identity()\n",
       "  (215): ParameterModule()\n",
       "  (216): ParameterModule()\n",
       "  (217): Identity()\n",
       "  (218): ParameterModule()\n",
       "  (219): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (220): MemoryEfficientSwish()\n",
       "  (221): Identity()\n",
       "  (222): ParameterModule()\n",
       "  (223): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (224): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (225): ParameterModule()\n",
       "  (226): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (227): Identity()\n",
       "  (228): ParameterModule()\n",
       "  (229): ParameterModule()\n",
       "  (230): Identity()\n",
       "  (231): ParameterModule()\n",
       "  (232): ParameterModule()\n",
       "  (233): Identity()\n",
       "  (234): ParameterModule()\n",
       "  (235): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (236): MemoryEfficientSwish()\n",
       "  (237): Identity()\n",
       "  (238): ParameterModule()\n",
       "  (239): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (240): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "  (241): ParameterModule()\n",
       "  (242): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (243): Identity()\n",
       "  (244): ParameterModule()\n",
       "  (245): ParameterModule()\n",
       "  (246): Identity()\n",
       "  (247): ParameterModule()\n",
       "  (248): ParameterModule()\n",
       "  (249): Identity()\n",
       "  (250): ParameterModule()\n",
       "  (251): BatchNorm2d(320, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (252): MemoryEfficientSwish()\n",
       "), Sequential(\n",
       "  (0): Identity()\n",
       "  (1): ParameterModule()\n",
       "  (2): BatchNorm2d(1280, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (3): AdaptiveAvgPool2d(output_size=1)\n",
       "  (4): Dropout(p=0.2, inplace=False)\n",
       "  (5): Linear(in_features=1280, out_features=50, bias=True)\n",
       "  (6): MemoryEfficientSwish()\n",
       ")], add_time=True, silent=False)\n",
       "loss_scale: 65536\n",
       "max_noskip: 1000\n",
       "dynamic: True\n",
       "clip: None\n",
       "flat_master: False\n",
       "max_scale: 16777216\n",
       "loss_fp32: True], layer_groups=[Sequential(\n",
       "  (0): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "  (1): ParameterModule()\n",
       "  (2): BatchNorm2d(32, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "), Sequential(\n",
       "  (0): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "  (1): ParameterModule()\n",
       "  (2): BatchNorm2d(32, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (3): Identity()\n",
       "  (4): ParameterModule()\n",
       "  (5): ParameterModule()\n",
       "  (6): Identity()\n",
       "  (7): ParameterModule()\n",
       "  (8): ParameterModule()\n",
       "  (9): Identity()\n",
       "  (10): ParameterModule()\n",
       "  (11): BatchNorm2d(16, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (12): MemoryEfficientSwish()\n",
       "  (13): Identity()\n",
       "  (14): ParameterModule()\n",
       "  (15): BatchNorm2d(96, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (16): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "  (17): ParameterModule()\n",
       "  (18): BatchNorm2d(96, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (19): Identity()\n",
       "  (20): ParameterModule()\n",
       "  (21): ParameterModule()\n",
       "  (22): Identity()\n",
       "  (23): ParameterModule()\n",
       "  (24): ParameterModule()\n",
       "  (25): Identity()\n",
       "  (26): ParameterModule()\n",
       "  (27): BatchNorm2d(24, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (28): MemoryEfficientSwish()\n",
       "  (29): Identity()\n",
       "  (30): ParameterModule()\n",
       "  (31): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (32): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "  (33): ParameterModule()\n",
       "  (34): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (35): Identity()\n",
       "  (36): ParameterModule()\n",
       "  (37): ParameterModule()\n",
       "  (38): Identity()\n",
       "  (39): ParameterModule()\n",
       "  (40): ParameterModule()\n",
       "  (41): Identity()\n",
       "  (42): ParameterModule()\n",
       "  (43): BatchNorm2d(24, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (44): MemoryEfficientSwish()\n",
       "  (45): Identity()\n",
       "  (46): ParameterModule()\n",
       "  (47): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (48): ZeroPad2d(padding=(1, 2, 1, 2), value=0.0)\n",
       "  (49): ParameterModule()\n",
       "  (50): BatchNorm2d(144, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (51): Identity()\n",
       "  (52): ParameterModule()\n",
       "  (53): ParameterModule()\n",
       "  (54): Identity()\n",
       "  (55): ParameterModule()\n",
       "  (56): ParameterModule()\n",
       "  (57): Identity()\n",
       "  (58): ParameterModule()\n",
       "  (59): BatchNorm2d(40, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (60): MemoryEfficientSwish()\n",
       "  (61): Identity()\n",
       "  (62): ParameterModule()\n",
       "  (63): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (64): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (65): ParameterModule()\n",
       "  (66): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (67): Identity()\n",
       "  (68): ParameterModule()\n",
       "  (69): ParameterModule()\n",
       "  (70): Identity()\n",
       "  (71): ParameterModule()\n",
       "  (72): ParameterModule()\n",
       "  (73): Identity()\n",
       "  (74): ParameterModule()\n",
       "  (75): BatchNorm2d(40, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (76): MemoryEfficientSwish()\n",
       "  (77): Identity()\n",
       "  (78): ParameterModule()\n",
       "  (79): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (80): ZeroPad2d(padding=(0, 1, 0, 1), value=0.0)\n",
       "  (81): ParameterModule()\n",
       "  (82): BatchNorm2d(240, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (83): Identity()\n",
       "  (84): ParameterModule()\n",
       "  (85): ParameterModule()\n",
       "  (86): Identity()\n",
       "  (87): ParameterModule()\n",
       "  (88): ParameterModule()\n",
       "  (89): Identity()\n",
       "  (90): ParameterModule()\n",
       "  (91): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (92): MemoryEfficientSwish()\n",
       "  (93): Identity()\n",
       "  (94): ParameterModule()\n",
       "  (95): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (96): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "  (97): ParameterModule()\n",
       "  (98): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (99): Identity()\n",
       "  (100): ParameterModule()\n",
       "  (101): ParameterModule()\n",
       "  (102): Identity()\n",
       "  (103): ParameterModule()\n",
       "  (104): ParameterModule()\n",
       "  (105): Identity()\n",
       "  (106): ParameterModule()\n",
       "  (107): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (108): MemoryEfficientSwish()\n",
       "  (109): Identity()\n",
       "  (110): ParameterModule()\n",
       "  (111): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (112): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "  (113): ParameterModule()\n",
       "  (114): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (115): Identity()\n",
       "  (116): ParameterModule()\n",
       "  (117): ParameterModule()\n",
       "  (118): Identity()\n",
       "  (119): ParameterModule()\n",
       "  (120): ParameterModule()\n",
       "  (121): Identity()\n",
       "  (122): ParameterModule()\n",
       "  (123): BatchNorm2d(80, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (124): MemoryEfficientSwish()\n",
       "  (125): Identity()\n",
       "  (126): ParameterModule()\n",
       "  (127): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (128): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (129): ParameterModule()\n",
       "  (130): BatchNorm2d(480, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (131): Identity()\n",
       "  (132): ParameterModule()\n",
       "  (133): ParameterModule()\n",
       "  (134): Identity()\n",
       "  (135): ParameterModule()\n",
       "  (136): ParameterModule()\n",
       "  (137): Identity()\n",
       "  (138): ParameterModule()\n",
       "  (139): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (140): MemoryEfficientSwish()\n",
       "  (141): Identity()\n",
       "  (142): ParameterModule()\n",
       "  (143): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (144): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (145): ParameterModule()\n",
       "  (146): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (147): Identity()\n",
       "  (148): ParameterModule()\n",
       "  (149): ParameterModule()\n",
       "  (150): Identity()\n",
       "  (151): ParameterModule()\n",
       "  (152): ParameterModule()\n",
       "  (153): Identity()\n",
       "  (154): ParameterModule()\n",
       "  (155): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (156): MemoryEfficientSwish()\n",
       "  (157): Identity()\n",
       "  (158): ParameterModule()\n",
       "  (159): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (160): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (161): ParameterModule()\n",
       "  (162): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (163): Identity()\n",
       "  (164): ParameterModule()\n",
       "  (165): ParameterModule()\n",
       "  (166): Identity()\n",
       "  (167): ParameterModule()\n",
       "  (168): ParameterModule()\n",
       "  (169): Identity()\n",
       "  (170): ParameterModule()\n",
       "  (171): BatchNorm2d(112, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (172): MemoryEfficientSwish()\n",
       "  (173): Identity()\n",
       "  (174): ParameterModule()\n",
       "  (175): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (176): ZeroPad2d(padding=(1, 2, 1, 2), value=0.0)\n",
       "  (177): ParameterModule()\n",
       "  (178): BatchNorm2d(672, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (179): Identity()\n",
       "  (180): ParameterModule()\n",
       "  (181): ParameterModule()\n",
       "  (182): Identity()\n",
       "  (183): ParameterModule()\n",
       "  (184): ParameterModule()\n",
       "  (185): Identity()\n",
       "  (186): ParameterModule()\n",
       "  (187): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (188): MemoryEfficientSwish()\n",
       "  (189): Identity()\n",
       "  (190): ParameterModule()\n",
       "  (191): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (192): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (193): ParameterModule()\n",
       "  (194): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (195): Identity()\n",
       "  (196): ParameterModule()\n",
       "  (197): ParameterModule()\n",
       "  (198): Identity()\n",
       "  (199): ParameterModule()\n",
       "  (200): ParameterModule()\n",
       "  (201): Identity()\n",
       "  (202): ParameterModule()\n",
       "  (203): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (204): MemoryEfficientSwish()\n",
       "  (205): Identity()\n",
       "  (206): ParameterModule()\n",
       "  (207): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (208): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (209): ParameterModule()\n",
       "  (210): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (211): Identity()\n",
       "  (212): ParameterModule()\n",
       "  (213): ParameterModule()\n",
       "  (214): Identity()\n",
       "  (215): ParameterModule()\n",
       "  (216): ParameterModule()\n",
       "  (217): Identity()\n",
       "  (218): ParameterModule()\n",
       "  (219): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (220): MemoryEfficientSwish()\n",
       "  (221): Identity()\n",
       "  (222): ParameterModule()\n",
       "  (223): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (224): ZeroPad2d(padding=(2, 2, 2, 2), value=0.0)\n",
       "  (225): ParameterModule()\n",
       "  (226): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (227): Identity()\n",
       "  (228): ParameterModule()\n",
       "  (229): ParameterModule()\n",
       "  (230): Identity()\n",
       "  (231): ParameterModule()\n",
       "  (232): ParameterModule()\n",
       "  (233): Identity()\n",
       "  (234): ParameterModule()\n",
       "  (235): BatchNorm2d(192, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (236): MemoryEfficientSwish()\n",
       "  (237): Identity()\n",
       "  (238): ParameterModule()\n",
       "  (239): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (240): ZeroPad2d(padding=(1, 1, 1, 1), value=0.0)\n",
       "  (241): ParameterModule()\n",
       "  (242): BatchNorm2d(1152, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (243): Identity()\n",
       "  (244): ParameterModule()\n",
       "  (245): ParameterModule()\n",
       "  (246): Identity()\n",
       "  (247): ParameterModule()\n",
       "  (248): ParameterModule()\n",
       "  (249): Identity()\n",
       "  (250): ParameterModule()\n",
       "  (251): BatchNorm2d(320, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (252): MemoryEfficientSwish()\n",
       "), Sequential(\n",
       "  (0): Identity()\n",
       "  (1): ParameterModule()\n",
       "  (2): BatchNorm2d(1280, eps=0.001, momentum=0.010000000000000009, affine=True, track_running_stats=True)\n",
       "  (3): AdaptiveAvgPool2d(output_size=1)\n",
       "  (4): Dropout(p=0.2, inplace=False)\n",
       "  (5): Linear(in_features=1280, out_features=50, bias=True)\n",
       "  (6): MemoryEfficientSwish()\n",
       ")], add_time=True, silent=False)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "arch = EfficientNet.from_pretrained('efficientnet-b0',num_classes=data.c)\n",
    "#arch = EfficientNet.from_pretrained('efficientnet-b3',num_classes=data.c)\n",
    "#arch = EfficientNet.from_pretrained('efficientnet-b7',num_classes=data.c)\n",
    "#loss_func = LabelSmoothingCrossEntropy()\n",
    "learn = Learner(data, arch, metrics = accuracy)\n",
    "learn = learn.split([learn.model._conv_stem,learn.model._blocks,learn.model._conv_head])\n",
    "learn.to_fp16()\n",
    "learn.lr_find()\n",
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "classes = learn.data.classes\n",
    "classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "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>2.800721</td>\n",
       "      <td>1.781202</td>\n",
       "      <td>0.522646</td>\n",
       "      <td>02:13</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(8,(1e-4, 1e-3, 1e-2), wd=0.2)\n",
    "learn.recorder.plot_losses()\n",
    "learn.recorder.plot_lr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.unfreeze()\n",
    "learn.lr_find()\n",
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.fit_one_cycle(12, max_lr=(1e-5,1e-4,1e-3), wd=(1e-4,1e-3,0.2))\n",
    "learn.recorder .plot_losses() \n",
    "learn.recorder.plot_lr()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.to_fp32()\n",
    "#preds, _ = learn.TTA(ds_type=DatasetType.Test)\n",
    "preds, _ = learn.get_preds(ds_type=DatasetType.Test)\n",
    "classes = learn.data.classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "scrolled": true
   },
   "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>id</th>\n",
       "      <th>label</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>313</td>\n",
       "      <td>[0.00071702315, 0.0014694027, 0.007375723, 0.0...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>138</td>\n",
       "      <td>[0.015893998, 0.0019469643, 0.00084006845, 0.0...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>256</td>\n",
       "      <td>[0.0014347282, 0.010931978, 0.001552489, 0.001...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>284</td>\n",
       "      <td>[0.0019938024, 0.0014186712, 0.044285625, 0.03...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>230</td>\n",
       "      <td>[0.00032279466, 0.0019879465, 0.00010182013, 0...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    id                                              label\n",
       "0  313  [0.00071702315, 0.0014694027, 0.007375723, 0.0...\n",
       "1  138  [0.015893998, 0.0019469643, 0.00084006845, 0.0...\n",
       "2  256  [0.0014347282, 0.010931978, 0.001552489, 0.001...\n",
       "3  284  [0.0019938024, 0.0014186712, 0.044285625, 0.03...\n",
       "4  230  [0.00032279466, 0.0019879465, 0.00010182013, 0..."
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results  = ['' for x in range(preds.shape[0])]\n",
    "for i in np.arange(preds.shape[0]):\n",
    "    #results[i] = preds[i].numpy()\n",
    "    results[i] = classes[np.argmax(preds[i]).numpy()]\n",
    "ids = [item.name[:-4] for item in learn.data.test_ds.items]\n",
    "df = pd.DataFrame({'id':ids, 'label': results})\n",
    "#df = pd.DataFrame(results, ids,columns=['id','label'])\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "fastai",
   "language": "python",
   "name": "fastai"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}