2-product-detection/[v8] Product Detection.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !pip install git+https://github.com/aleju/imgaug.git git+https://github.com/docPhil99/ImageDataAugmentor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "import tensorflow.keras as tfk\n",
    "import tensorflow_addons as tfa\n",
    "\n",
    "from tensorflow.keras.preprocessing import image as tfk_image\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator as tfk_ImageDataGenerator\n",
    "\n",
    "#from tensorflow.python.keras.applications.efficientnet import EfficientNetB4 as tfk_BaseModel\n",
    "#from tensorflow.python.keras.applications.efficientnet import preprocess_input as tfk_preprocess_input\n",
    "\n",
    "#from tensorflow.python.keras.applications.inception_v3 import InceptionV3 as tfk_BaseModel\n",
    "#from tensorflow.python.keras.applications.inception_v3 import preprocess_input as tfk_preprocess_input\n",
    "\n",
    "#from tensorflow.python.keras.applications.resnet import ResNet50 as tfk_BaseModel\n",
    "#from tensorflow.python.keras.applications.resnet import preprocess_input as tfk_preprocess_input\n",
    "\n",
    "from tensorflow.python.keras.applications.xception import Xception as tfk_BaseModel\n",
    "from tensorflow.python.keras.applications.xception import preprocess_input as tfk_preprocess_input\n",
    "\n",
    "from ImageDataAugmentor.image_data_augmentor import ImageDataAugmentor\n",
    "import albumentations as A\n",
    "\n",
    "import datetime, os\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns; sns.set();\n",
    "from PIL import Image, ImageEnhance\n",
    "\n",
    "from sklearn.utils import class_weight\n",
    "from sklearn.metrics import confusion_matrix, log_loss, accuracy_score\n",
    "from sklearn.preprocessing import OneHotEncoder"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Prepare Dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load and combine datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "pd_shopee_train = pd.read_csv('shopee-product-detection-dataset/train.csv')\n",
    "# pd_confusing_images = pd.read_csv('shopee-product-detection-dataset/confusing_images_2__mistake_ratio=0.9.csv')\n",
    "pd_amazon_train = pd.read_csv('amazon-scraped-dataset/amazon.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# uncomment to clean up the shopee dataset\n",
    "# mask = ~pd_shopee_train.set_index('dir_filename').index.isin(pd_confusing_images.set_index('dir_filename').index)\n",
    "# pd_shopee_train = pd_shopee_train.loc[mask].copy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "pd_shopee_train = pd_shopee_train.sample(frac=1).reset_index(drop=True)\n",
    "pd_shopee_train['category']     = pd_shopee_train['category'].apply(lambda x : '{:02d}'.format(x))\n",
    "pd_shopee_train['dir_filename'] = pd_shopee_train.apply(lambda x: 'shopee-product-detection-dataset/train/{}/{}'.format(x.category, x.filename), axis=1)\n",
    "\n",
    "pd_amazon_train = pd_amazon_train.sample(frac=1).reset_index(drop=True)\n",
    "pd_amazon_train['category']     = pd_amazon_train['category'].apply(lambda x : '{:02d}'.format(x))\n",
    "pd_amazon_train['dir_filename'] = pd_amazon_train.apply(lambda x: 'amazon-scraped-dataset/{}/{}'.format(x.category, x.filename), axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "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>filename</th>\n",
       "      <th>category</th>\n",
       "      <th>dir_filename</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>dcc1e336342fc2da16c7b6411660adbf.jpg</td>\n",
       "      <td>19</td>\n",
       "      <td>shopee-product-detection-dataset/train/19/dcc1...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>63927ec859cf4c6afe452841dafa47ef.jpg</td>\n",
       "      <td>39</td>\n",
       "      <td>shopee-product-detection-dataset/train/39/6392...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>f2aef279f4e18fcafd2d52d3004dbcd9.jpg</td>\n",
       "      <td>09</td>\n",
       "      <td>shopee-product-detection-dataset/train/09/f2ae...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>d2f4445f1adf32d10551f7978614af87.jpg</td>\n",
       "      <td>34</td>\n",
       "      <td>shopee-product-detection-dataset/train/34/d2f4...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>3d87bc36a317e930b84013530643019f.jpg</td>\n",
       "      <td>29</td>\n",
       "      <td>shopee-product-detection-dataset/train/29/3d87...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                               filename category  \\\n",
       "0  dcc1e336342fc2da16c7b6411660adbf.jpg       19   \n",
       "1  63927ec859cf4c6afe452841dafa47ef.jpg       39   \n",
       "2  f2aef279f4e18fcafd2d52d3004dbcd9.jpg       09   \n",
       "3  d2f4445f1adf32d10551f7978614af87.jpg       34   \n",
       "4  3d87bc36a317e930b84013530643019f.jpg       29   \n",
       "\n",
       "                                        dir_filename  \n",
       "0  shopee-product-detection-dataset/train/19/dcc1...  \n",
       "1  shopee-product-detection-dataset/train/39/6392...  \n",
       "2  shopee-product-detection-dataset/train/09/f2ae...  \n",
       "3  shopee-product-detection-dataset/train/34/d2f4...  \n",
       "4  shopee-product-detection-dataset/train/29/3d87...  "
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd_combined_dataset = pd_shopee_train.merge(pd_amazon_train, how='outer')\n",
    "pd_combined_dataset.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "assert(len(pd_shopee_train) + len(pd_amazon_train) == len(pd_combined_dataset))\n",
    "assert(np.all(np.unique(pd_combined_dataset['category']) == \n",
    "              np.array(['{:02d}'.format(cat) for cat in range(42)])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Prepare data generators"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_shape = (256, 256, 3)\n",
    "batch_size  = 64 #64+32"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "AUGMENTATIONS = A.Compose([\n",
    "    A.HorizontalFlip(),\n",
    "    A.VerticalFlip(),\n",
    "    A.Rotate(),\n",
    "    A.OneOf([\n",
    "        A.RandomContrast(),\n",
    "        A.RandomGamma(),\n",
    "        A.RandomBrightness(),\n",
    "    ], p = 0.3),\n",
    "    A.OneOf([\n",
    "        A.ElasticTransform(\n",
    "            alpha = 120,\n",
    "            sigma = 120 * 0.05,\n",
    "            alpha_affine = 120 * 0.03\n",
    "        ),\n",
    "        A.GridDistortion(),\n",
    "        A.OpticalDistortion(\n",
    "            distort_limit = 2,\n",
    "            shift_limit = 0.5\n",
    "        ),\n",
    "    ], p = 0.3),\n",
    "    A.OneOf([\n",
    "        A.Cutout(\n",
    "            num_holes=4,\n",
    "            max_h_size=64,\n",
    "            max_w_size=64,\n",
    "            #fill_mode=\"gaussian\"\n",
    "        ),\n",
    "        A.GridDropout(\n",
    "            ratio=0.25,\n",
    "            holes_number_x=8,\n",
    "            holes_number_y=8,\n",
    "            #fill_mode=\"gaussian\"\n",
    "        ),\n",
    "    ], p=1),\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 94554 validated image filenames belonging to 42 classes.\n",
      "Found 10505 validated image filenames belonging to 42 classes.\n"
     ]
    }
   ],
   "source": [
    "train_datagen = ImageDataAugmentor(\n",
    "    preprocess_input=tfk_preprocess_input,\n",
    "    validation_split=0.1,\n",
    "    augment=AUGMENTATIONS,\n",
    ")\n",
    "\n",
    "train_generator = train_datagen.flow_from_dataframe(\n",
    "    pd_shopee_train,\n",
    "    directory='',\n",
    "    x_col='dir_filename',\n",
    "    y_col='category',\n",
    "    target_size=input_shape[:2],\n",
    "    color_mode='rgb',\n",
    "    batch_size=batch_size,\n",
    "    class_mode='categorical',\n",
    "    shuffle=True,\n",
    "    subset='training'\n",
    ")\n",
    "\n",
    "val_generator = train_datagen.flow_from_dataframe(\n",
    "    pd_shopee_train,\n",
    "    directory='',\n",
    "    x_col='dir_filename',\n",
    "    y_col='category',\n",
    "    target_size=input_shape[:2],\n",
    "    color_mode='rgb',\n",
    "    batch_size=batch_size,\n",
    "    class_mode='categorical',\n",
    "    shuffle=True,\n",
    "    subset='validation'\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Check class distributions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.title(\"Class Distribution in Combined Dataset\")\n",
    "plt.xlabel(\"Class\")\n",
    "plt.ylabel(\"Freq\")\n",
    "plt.hist(pd_combined_dataset['category'].astype(int), bins=np.arange(43));"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(train_generator.classes, bins=np.arange(43));"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(val_generator.classes, bins=np.arange(43));"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Calculate class weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "class_weights = class_weight.compute_class_weight('balanced',\n",
    "                                                  np.unique(train_generator.classes),\n",
    "                                                  train_generator.classes)\n",
    "class_weights = {i : class_weights[i] for i in range(len(class_weights))}\n",
    "#for i in [0, 1, 2, 3, 20, 41]:\n",
    "#    class_weights[i] = 1.25\n",
    "plt.title(\"Class Weights\")\n",
    "plt.ylabel(\"Weight\")\n",
    "plt.xlabel(\"Class\")\n",
    "plt.bar(class_weights.keys(), class_weights.values());"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Prepare Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "base_model = tfk_BaseModel(input_shape=input_shape, weights='imagenet', include_top=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "TYPE = 2\n",
    "if TYPE == 1:\n",
    "    for layer in base_model.layers[:-18]:\n",
    "        layer.trainable = False\n",
    "    model = tfk.Sequential([\n",
    "        base_model,\n",
    "        tfk.layers.GlobalMaxPooling2D(),\n",
    "        tfk.layers.Dense(512, activation='relu'),\n",
    "        tfk.layers.Dense(128, activation='relu'),\n",
    "        tfk.layers.Dense(42, activation='softmax', name='output_layer'),\n",
    "    ])\n",
    "elif TYPE == 2:\n",
    "    model = tfk.Sequential([\n",
    "        base_model,\n",
    "        tfk.layers.GlobalMaxPooling2D(),\n",
    "        tfk.layers.Dropout(rate=0.2),\n",
    "        tfk.layers.Dense(42, activation='softmax', name='output_layer'),\n",
    "    ])\n",
    "elif TYPE == 3:\n",
    "    for layer in base_model.layers[:-len(base_model.layers)//5]:\n",
    "        layer.trainable = False\n",
    "    model = tfk.Sequential([\n",
    "        base_model,\n",
    "        tfk.layers.GlobalMaxPooling2D(),\n",
    "        tfk.layers.Dropout(rate=0.2),\n",
    "        tfk.layers.Dense(42, activation='softmax', name='output_layer'),\n",
    "    ])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "xception (Model)             (None, 8, 8, 2048)        20861480  \n",
      "_________________________________________________________________\n",
      "global_max_pooling2d (Global (None, 2048)              0         \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, 2048)              0         \n",
      "_________________________________________________________________\n",
      "output_layer (Dense)         (None, 42)                86058     \n",
      "=================================================================\n",
      "Total params: 20,947,538\n",
      "Trainable params: 20,893,010\n",
      "Non-trainable params: 54,528\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "tfk.utils.plot_model(model, rankdir='LR', dpi=300, show_shapes=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model.save('model_aug_full_4.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train Top Layers of Model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model = tfk.models.load_model('model_aug_full_cleaner_19.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Prepare callbacks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "reduce_lr = tf.keras.callbacks.ReduceLROnPlateau(\n",
    "    monitor='val_loss',\n",
    "    factor=0.2,\n",
    "    patience=10,\n",
    "    min_lr=1e-7\n",
    ")\n",
    "\n",
    "early_stop = tfk.callbacks.EarlyStopping(\n",
    "    monitor='val_loss',\n",
    "    patience=30,\n",
    "    restore_best_weights=True\n",
    ")\n",
    "\n",
    "log_dir = os.path.join(\".model_files/logs/\" + datetime.datetime.now().strftime(\"%Y%m%d-%H%M%S\"))\n",
    "if not os.path.exists(log_dir):\n",
    "    os.makedirs(log_dir)\n",
    "\n",
    "checkpoint = tfk.callbacks.ModelCheckpoint(\n",
    "    filepath = log_dir + '/ckpt/',\n",
    "    save_freq ='epoch',\n",
    "    save_best_only=True,\n",
    "    monitor='val_loss'\n",
    ")\n",
    "\n",
    "tensorboard_callback = tf.keras.callbacks.TensorBoard(\n",
    "    log_dir=log_dir,\n",
    "    histogram_freq=0,\n",
    "    update_freq=100,\n",
    "    write_graph=False,\n",
    "    profile_batch = 0\n",
    ")\n",
    "\n",
    "class LRTensorBoard(tf.keras.callbacks.TensorBoard):\n",
    "    def __init__(self, log_dir, **kwargs):  # add other arguments to __init__ if you need\n",
    "        super().__init__(log_dir=log_dir, **kwargs)\n",
    "\n",
    "    def on_epoch_end(self, epoch, logs=None):\n",
    "        logs = logs or {}\n",
    "        lr_sched = getattr(self.model.optimizer, \"lr\", None)\n",
    "        logs.update({'lr': tf.keras.backend.get_value(lr_sched(tf.cast(self.model.optimizer.iterations, tf.float32)))})\n",
    "        super().on_epoch_end(epoch, logs)\n",
    "\n",
    "lr_tensorboard_callback = LRTensorBoard(log_dir=log_dir)\n",
    "        \n",
    "#tqdm_callback = tfa.callbacks.TQDMCallback()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Compile model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr_schedule = tfa.optimizers.ExponentialCyclicalLearningRate(\n",
    "    initial_learning_rate=1e-6,\n",
    "    maximal_learning_rate=1e-3,\n",
    "    step_size=2*train_generator.n//batch_size\n",
    ")\n",
    "\n",
    "model.compile(optimizer=tfk.optimizers.Adam(learning_rate=5e-5),\n",
    "              loss=tfk.losses.CategoricalCrossentropy(label_smoothing=0.1),\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "histories = []"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for v in range(26, 30+1):\n",
    "    print(\"---------------- {} ----------------\".format(v))\n",
    "    print(\"Learning rate: {}\".format(tfk.backend.get_value(getattr(model.optimizer, \"lr\", None))))\n",
    "    history = model.fit(train_generator,\n",
    "                        initial_epoch=0,\n",
    "                        epochs=15,\n",
    "                        steps_per_epoch=200,\n",
    "                        validation_steps=20,\n",
    "                        validation_data=val_generator,\n",
    "                        class_weight=class_weights,\n",
    "                        callbacks=[reduce_lr,\n",
    "                                   early_stop, \n",
    "                                   #checkpoint, \n",
    "                                   tensorboard_callback,\n",
    "                                   #lr_tensorboard_callback,\n",
    "                                   #tqdm_callback\n",
    "                                  ]\n",
    "                       )\n",
    "    histories.append(pd.DataFrame(history.history))\n",
    "    pd.concat(histories).to_csv(\"history_aug_full_cleaner_{}.csv\".format(v), index=False)\n",
    "    model.save('model_aug_full_cleaner_{}.h5'.format(v))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Testing"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We'll do in-sample testing here. A bit non-standard, but good enough."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model = tfk.models.load_model('model_clean_3.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Prepare unaugmented dataset generator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 105059 validated image filenames belonging to 42 classes.\n"
     ]
    }
   ],
   "source": [
    "clean_datagen = tfk_ImageDataGenerator(\n",
    "    preprocessing_function=tfk_preprocess_input,\n",
    ")\n",
    "\n",
    "clean_generator = clean_datagen.flow_from_dataframe(\n",
    "    pd_shopee_train,\n",
    "    directory='',\n",
    "    x_col='dir_filename',\n",
    "    y_col='category',\n",
    "    target_size=input_shape[:2],\n",
    "    color_mode='rgb',\n",
    "    batch_size=batch_size,\n",
    "    class_mode='categorical',\n",
    "    shuffle=True,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### In-sample Testing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "step: 10\n",
      "step: 20\n",
      "step: 30\n",
      "step: 40\n",
      "step: 50\n",
      "Wall time: 23.2 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "y_pred = []\n",
    "y_true = []\n",
    "\n",
    "for i in range(50):\n",
    "    if (i+1) % 10 == 0:\n",
    "        print(\"step:\", i+1)\n",
    "    x, y = next(iter(clean_generator))\n",
    "    y_true.extend(y)\n",
    "    y_pred.extend(model.predict(x))\n",
    "\n",
    "y_pred = np.array(y_pred)\n",
    "y_true = np.array(y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "matrix = confusion_matrix(y_true.argmax(axis=1), y_pred.argmax(axis=1))\n",
    "matrix_ = matrix.copy()\n",
    "for i in range(len(matrix)):\n",
    "    matrix_[i, i] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x648 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize = (16,9))\n",
    "sns.heatmap(matrix_, linecolor='grey', linewidth=0.1)\n",
    "plt.title(\"Confusion Matrix\")\n",
    "plt.xlabel(\"Predicted Label\")\n",
    "plt.ylabel(\"True Label\")\n",
    "plt.savefig('confusion_matrix.png', \n",
    "            #transparent=True,\n",
    "            bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [],
   "source": [
    "def label_smoother(x, alpha=0.1):\n",
    "    return (1 - alpha) * x + alpha / x.shape[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loss: 1.1015037\n",
      "Accuracy: 0.9475\n"
     ]
    }
   ],
   "source": [
    "loss     = tfk.losses.categorical_crossentropy(y_pred, label_smoother(y_true, alpha=0.1)).numpy().mean()\n",
    "accuracy = accuracy_score(y_pred.argmax(axis=1), y_true.argmax(axis=1))\n",
    "\n",
    "print(\"Loss:\", loss)\n",
    "print(\"Accuracy:\", accuracy)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Generate Submission"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Prepare test data generator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "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>filename</th>\n",
       "      <th>category</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>fd663cf2b6e1d7b02938c6aaae0a32d2.jpg</td>\n",
       "      <td>43</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>c7fd77508a8c355eaab0d4e10efd6b15.jpg</td>\n",
       "      <td>43</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>127f3e6d6e3491b2459812353f33a913.jpg</td>\n",
       "      <td>43</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>5ca4f2da11eda083064e6c36f37eeb81.jpg</td>\n",
       "      <td>43</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>46d681a542f2c71be017eef6aae23313.jpg</td>\n",
       "      <td>43</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                               filename  category\n",
       "0  fd663cf2b6e1d7b02938c6aaae0a32d2.jpg        43\n",
       "1  c7fd77508a8c355eaab0d4e10efd6b15.jpg        43\n",
       "2  127f3e6d6e3491b2459812353f33a913.jpg        43\n",
       "3  5ca4f2da11eda083064e6c36f37eeb81.jpg        43\n",
       "4  46d681a542f2c71be017eef6aae23313.jpg        43"
      ]
     },
     "execution_count": 74,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd_test = pd.read_csv('shopee-product-detection-dataset/test.csv')\n",
    "pd_test.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 12186 validated image filenames.\n"
     ]
    }
   ],
   "source": [
    "test_datagen = tfk_ImageDataGenerator(\n",
    "    preprocessing_function=tfk_preprocess_input\n",
    ")\n",
    "\n",
    "test_generator = test_datagen.flow_from_dataframe(\n",
    "    pd_test,\n",
    "    directory='shopee-product-detection-dataset/test/',\n",
    "    x_col='filename',\n",
    "    target_size=input_shape[:2],\n",
    "    color_mode='rgb',\n",
    "    batch_size=batch_size,\n",
    "    class_mode=None,\n",
    "    shuffle=False,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Predict categories"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "191/191 [==============================] - 57s 299ms/step\n"
     ]
    }
   ],
   "source": [
    "y_pred = model.predict(test_generator, verbose=1, steps=test_generator.n//batch_size+1)\n",
    "y_pred = y_pred[:test_generator.n]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "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>filename</th>\n",
       "      <th>category</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>fd663cf2b6e1d7b02938c6aaae0a32d2.jpg</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>c7fd77508a8c355eaab0d4e10efd6b15.jpg</td>\n",
       "      <td>27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>127f3e6d6e3491b2459812353f33a913.jpg</td>\n",
       "      <td>04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>5ca4f2da11eda083064e6c36f37eeb81.jpg</td>\n",
       "      <td>22</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>46d681a542f2c71be017eef6aae23313.jpg</td>\n",
       "      <td>12</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                               filename category\n",
       "0  fd663cf2b6e1d7b02938c6aaae0a32d2.jpg       20\n",
       "1  c7fd77508a8c355eaab0d4e10efd6b15.jpg       27\n",
       "2  127f3e6d6e3491b2459812353f33a913.jpg       04\n",
       "3  5ca4f2da11eda083064e6c36f37eeb81.jpg       22\n",
       "4  46d681a542f2c71be017eef6aae23313.jpg       12"
      ]
     },
     "execution_count": 77,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd_submission = pd_test.copy()\n",
    "pd_submission['category'] = list(map(lambda x : '{:02d}'.format(x), y_pred.argmax(axis=1)))\n",
    "pd_submission.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "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>filename</th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "      <th>2</th>\n",
       "      <th>3</th>\n",
       "      <th>4</th>\n",
       "      <th>5</th>\n",
       "      <th>6</th>\n",
       "      <th>7</th>\n",
       "      <th>8</th>\n",
       "      <th>...</th>\n",
       "      <th>32</th>\n",
       "      <th>33</th>\n",
       "      <th>34</th>\n",
       "      <th>35</th>\n",
       "      <th>36</th>\n",
       "      <th>37</th>\n",
       "      <th>38</th>\n",
       "      <th>39</th>\n",
       "      <th>40</th>\n",
       "      <th>41</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>fd663cf2b6e1d7b02938c6aaae0a32d2.jpg</td>\n",
       "      <td>0.001062</td>\n",
       "      <td>0.000684</td>\n",
       "      <td>0.000134</td>\n",
       "      <td>0.000199</td>\n",
       "      <td>0.000540</td>\n",
       "      <td>0.000284</td>\n",
       "      <td>0.000295</td>\n",
       "      <td>0.000402</td>\n",
       "      <td>0.000438</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000095</td>\n",
       "      <td>0.000192</td>\n",
       "      <td>0.000609</td>\n",
       "      <td>0.000310</td>\n",
       "      <td>0.002228</td>\n",
       "      <td>0.000379</td>\n",
       "      <td>0.001500</td>\n",
       "      <td>0.001930</td>\n",
       "      <td>0.000345</td>\n",
       "      <td>0.051738</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>c7fd77508a8c355eaab0d4e10efd6b15.jpg</td>\n",
       "      <td>0.001451</td>\n",
       "      <td>0.001337</td>\n",
       "      <td>0.001185</td>\n",
       "      <td>0.000925</td>\n",
       "      <td>0.001583</td>\n",
       "      <td>0.001145</td>\n",
       "      <td>0.001150</td>\n",
       "      <td>0.001464</td>\n",
       "      <td>0.002079</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000806</td>\n",
       "      <td>0.000773</td>\n",
       "      <td>0.001577</td>\n",
       "      <td>0.001187</td>\n",
       "      <td>0.000801</td>\n",
       "      <td>0.000818</td>\n",
       "      <td>0.000771</td>\n",
       "      <td>0.001410</td>\n",
       "      <td>0.000974</td>\n",
       "      <td>0.000545</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>127f3e6d6e3491b2459812353f33a913.jpg</td>\n",
       "      <td>0.001669</td>\n",
       "      <td>0.001508</td>\n",
       "      <td>0.002610</td>\n",
       "      <td>0.001780</td>\n",
       "      <td>0.900689</td>\n",
       "      <td>0.001427</td>\n",
       "      <td>0.001080</td>\n",
       "      <td>0.002460</td>\n",
       "      <td>0.003593</td>\n",
       "      <td>...</td>\n",
       "      <td>0.001229</td>\n",
       "      <td>0.001360</td>\n",
       "      <td>0.001246</td>\n",
       "      <td>0.001766</td>\n",
       "      <td>0.000828</td>\n",
       "      <td>0.001636</td>\n",
       "      <td>0.003009</td>\n",
       "      <td>0.001626</td>\n",
       "      <td>0.001489</td>\n",
       "      <td>0.004999</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>5ca4f2da11eda083064e6c36f37eeb81.jpg</td>\n",
       "      <td>0.001432</td>\n",
       "      <td>0.001434</td>\n",
       "      <td>0.001345</td>\n",
       "      <td>0.000857</td>\n",
       "      <td>0.001710</td>\n",
       "      <td>0.000863</td>\n",
       "      <td>0.001031</td>\n",
       "      <td>0.000886</td>\n",
       "      <td>0.001776</td>\n",
       "      <td>...</td>\n",
       "      <td>0.001504</td>\n",
       "      <td>0.001558</td>\n",
       "      <td>0.003291</td>\n",
       "      <td>0.001593</td>\n",
       "      <td>0.001262</td>\n",
       "      <td>0.001746</td>\n",
       "      <td>0.001101</td>\n",
       "      <td>0.001213</td>\n",
       "      <td>0.000730</td>\n",
       "      <td>0.000641</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>46d681a542f2c71be017eef6aae23313.jpg</td>\n",
       "      <td>0.002208</td>\n",
       "      <td>0.001831</td>\n",
       "      <td>0.001414</td>\n",
       "      <td>0.001491</td>\n",
       "      <td>0.003105</td>\n",
       "      <td>0.002208</td>\n",
       "      <td>0.002535</td>\n",
       "      <td>0.001520</td>\n",
       "      <td>0.001259</td>\n",
       "      <td>...</td>\n",
       "      <td>0.001109</td>\n",
       "      <td>0.002540</td>\n",
       "      <td>0.001707</td>\n",
       "      <td>0.006560</td>\n",
       "      <td>0.000566</td>\n",
       "      <td>0.001022</td>\n",
       "      <td>0.001550</td>\n",
       "      <td>0.004861</td>\n",
       "      <td>0.001709</td>\n",
       "      <td>0.001399</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 43 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                               filename         0         1         2  \\\n",
       "0  fd663cf2b6e1d7b02938c6aaae0a32d2.jpg  0.001062  0.000684  0.000134   \n",
       "1  c7fd77508a8c355eaab0d4e10efd6b15.jpg  0.001451  0.001337  0.001185   \n",
       "2  127f3e6d6e3491b2459812353f33a913.jpg  0.001669  0.001508  0.002610   \n",
       "3  5ca4f2da11eda083064e6c36f37eeb81.jpg  0.001432  0.001434  0.001345   \n",
       "4  46d681a542f2c71be017eef6aae23313.jpg  0.002208  0.001831  0.001414   \n",
       "\n",
       "          3         4         5         6         7         8  ...        32  \\\n",
       "0  0.000199  0.000540  0.000284  0.000295  0.000402  0.000438  ...  0.000095   \n",
       "1  0.000925  0.001583  0.001145  0.001150  0.001464  0.002079  ...  0.000806   \n",
       "2  0.001780  0.900689  0.001427  0.001080  0.002460  0.003593  ...  0.001229   \n",
       "3  0.000857  0.001710  0.000863  0.001031  0.000886  0.001776  ...  0.001504   \n",
       "4  0.001491  0.003105  0.002208  0.002535  0.001520  0.001259  ...  0.001109   \n",
       "\n",
       "         33        34        35        36        37        38        39  \\\n",
       "0  0.000192  0.000609  0.000310  0.002228  0.000379  0.001500  0.001930   \n",
       "1  0.000773  0.001577  0.001187  0.000801  0.000818  0.000771  0.001410   \n",
       "2  0.001360  0.001246  0.001766  0.000828  0.001636  0.003009  0.001626   \n",
       "3  0.001558  0.003291  0.001593  0.001262  0.001746  0.001101  0.001213   \n",
       "4  0.002540  0.001707  0.006560  0.000566  0.001022  0.001550  0.004861   \n",
       "\n",
       "         40        41  \n",
       "0  0.000345  0.051738  \n",
       "1  0.000974  0.000545  \n",
       "2  0.001489  0.004999  \n",
       "3  0.000730  0.000641  \n",
       "4  0.001709  0.001399  \n",
       "\n",
       "[5 rows x 43 columns]"
      ]
     },
     "execution_count": 78,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd_submission_ = pd_submission.drop(['category'], axis=1)\n",
    "for c in range(42):\n",
    "    pd_submission_[c] = y_pred[:,c]\n",
    "pd_submission_.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {},
   "outputs": [],
   "source": [
    "version = 28\n",
    "pd_submission.to_csv('submission_{}.csv'.format(version), index=False)\n",
    "pd_submission_.to_csv('submission_{}_dist.csv'.format(version), index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Compare to previous submission"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "567 12186 0.0465288035450517\n"
     ]
    },
    {
     "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>filename</th>\n",
       "      <th>category_27</th>\n",
       "      <th>category</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>2e9521b6c76dd1d7bebbeb8a2d84cacc.jpg</td>\n",
       "      <td>14</td>\n",
       "      <td>38</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>33</th>\n",
       "      <td>26ef13b0da02bfa1ba211ee523b3fdbd.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>08</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>88</th>\n",
       "      <td>b00ec65ae02183e69f16e5a46d9d7973.jpg</td>\n",
       "      <td>26</td>\n",
       "      <td>19</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>91</th>\n",
       "      <td>5f6da1149fd4109457d375d70a8def4c.jpg</td>\n",
       "      <td>38</td>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>92</th>\n",
       "      <td>0f76078f68b2d6ae47e86ae488b6de08.jpg</td>\n",
       "      <td>00</td>\n",
       "      <td>02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>123</th>\n",
       "      <td>835de04e18d0eff0774584a555e42a2a.jpg</td>\n",
       "      <td>33</td>\n",
       "      <td>35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>159</th>\n",
       "      <td>805f92f1ddbd068096384d595596ffb9.jpg</td>\n",
       "      <td>12</td>\n",
       "      <td>34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>223</th>\n",
       "      <td>2d5c4d3c9dcae43e0317e985e339db54.jpg</td>\n",
       "      <td>02</td>\n",
       "      <td>00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>237</th>\n",
       "      <td>aca99804062a29a553c037c164a8c59b.jpg</td>\n",
       "      <td>38</td>\n",
       "      <td>07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>243</th>\n",
       "      <td>3a3aa10a8b3640139ffaa19437509b2e.jpg</td>\n",
       "      <td>31</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>276</th>\n",
       "      <td>bc50bb2ae42f5e9552815a4dff3ead59.jpg</td>\n",
       "      <td>09</td>\n",
       "      <td>36</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>294</th>\n",
       "      <td>97e9a171bd147edbb1feb62e941c356b.jpg</td>\n",
       "      <td>36</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>319</th>\n",
       "      <td>95b805c78c3f0b8ab7d2cd8cdb399000.jpg</td>\n",
       "      <td>07</td>\n",
       "      <td>22</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>335</th>\n",
       "      <td>a1b724d220b55fd36251fda8a663973a.jpg</td>\n",
       "      <td>22</td>\n",
       "      <td>03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>337</th>\n",
       "      <td>81211c018458dbf7dcd4312eda39d7e1.jpg</td>\n",
       "      <td>38</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>383</th>\n",
       "      <td>d340396a724ee484e81ca138bdd19e9d.jpg</td>\n",
       "      <td>32</td>\n",
       "      <td>22</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>394</th>\n",
       "      <td>4f47c0c0ec8882b2b374918b8a034e26.jpg</td>\n",
       "      <td>36</td>\n",
       "      <td>35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>408</th>\n",
       "      <td>098637c7d594d4779eaba0c3e7125648.jpg</td>\n",
       "      <td>30</td>\n",
       "      <td>16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>415</th>\n",
       "      <td>7dea637087ae033d0b95c4242fd83844.jpg</td>\n",
       "      <td>36</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>440</th>\n",
       "      <td>25499b4fc34655c01dca6cc0fd77602d.jpg</td>\n",
       "      <td>19</td>\n",
       "      <td>18</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                 filename category_27 category\n",
       "27   2e9521b6c76dd1d7bebbeb8a2d84cacc.jpg          14       38\n",
       "33   26ef13b0da02bfa1ba211ee523b3fdbd.jpg          41       08\n",
       "88   b00ec65ae02183e69f16e5a46d9d7973.jpg          26       19\n",
       "91   5f6da1149fd4109457d375d70a8def4c.jpg          38       40\n",
       "92   0f76078f68b2d6ae47e86ae488b6de08.jpg          00       02\n",
       "123  835de04e18d0eff0774584a555e42a2a.jpg          33       35\n",
       "159  805f92f1ddbd068096384d595596ffb9.jpg          12       34\n",
       "223  2d5c4d3c9dcae43e0317e985e339db54.jpg          02       00\n",
       "237  aca99804062a29a553c037c164a8c59b.jpg          38       07\n",
       "243  3a3aa10a8b3640139ffaa19437509b2e.jpg          31       20\n",
       "276  bc50bb2ae42f5e9552815a4dff3ead59.jpg          09       36\n",
       "294  97e9a171bd147edbb1feb62e941c356b.jpg          36       20\n",
       "319  95b805c78c3f0b8ab7d2cd8cdb399000.jpg          07       22\n",
       "335  a1b724d220b55fd36251fda8a663973a.jpg          22       03\n",
       "337  81211c018458dbf7dcd4312eda39d7e1.jpg          38       16\n",
       "383  d340396a724ee484e81ca138bdd19e9d.jpg          32       22\n",
       "394  4f47c0c0ec8882b2b374918b8a034e26.jpg          36       35\n",
       "408  098637c7d594d4779eaba0c3e7125648.jpg          30       16\n",
       "415  7dea637087ae033d0b95c4242fd83844.jpg          36       41\n",
       "440  25499b4fc34655c01dca6cc0fd77602d.jpg          19       18"
      ]
     },
     "execution_count": 80,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "now  = version\n",
    "cat_now_name  = 'category_{}'.format(now)\n",
    "prev = '27'\n",
    "cat_prev_name = 'category_{}'.format(prev)\n",
    "\n",
    "pd_submission_prev = pd.read_csv('submission_{}.csv'.format(prev))\n",
    "pd_submission_prev['category'] = pd_submission_prev['category'].apply(lambda x : '{:02d}'.format(x))\n",
    "pd_submission_prev.columns = ['filename', cat_prev_name]\n",
    "pd_submission_prev = pd_submission_prev.merge(pd_submission.sort_values(by='filename'), on='filename')\n",
    "pd_iba = pd_submission_prev[pd_submission_prev[cat_prev_name] != pd_submission_prev['category']]\n",
    "print(len(pd_iba), len(pd_submission_prev), len(pd_iba) / len(pd_submission_prev))\n",
    "pd_iba.head(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {},
   "outputs": [],
   "source": [
    "matrix = confusion_matrix(pd_submission_prev['category'], pd_submission_prev[cat_prev_name])\n",
    "matrix_ = matrix.copy()\n",
    "for i in range(len(matrix)):\n",
    "    matrix_[i, i] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x648 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize = (16,9))\n",
    "sns.heatmap(matrix_, linecolor='grey', linewidth=0.1)\n",
    "plt.title(\"Confusion Matrix\")\n",
    "plt.xlabel(\"Better Label\")\n",
    "plt.ylabel(\"Bad Label\")\n",
    "plt.savefig('confusion_matrix.png', \n",
    "            #transparent=True,\n",
    "            bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [],
   "source": [
    "pd_iba.to_csv(\"iba.csv\", index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "9 0.0007385524372230429\n"
     ]
    },
    {
     "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>filename</th>\n",
       "      <th>category_27</th>\n",
       "      <th>category</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1952</th>\n",
       "      <td>a3ebc60c3f99cbdb957c0e66ce542661.jpg</td>\n",
       "      <td>20</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3134</th>\n",
       "      <td>5b62c3aed0dcbfb0fea50041e9a232c8.jpg</td>\n",
       "      <td>20</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3791</th>\n",
       "      <td>4f4ac25f31a06933a6d10f127b3a7778.jpg</td>\n",
       "      <td>20</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6162</th>\n",
       "      <td>a6ffb9e99b156aed038f4023c458a279.jpg</td>\n",
       "      <td>20</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8019</th>\n",
       "      <td>d8d36c64fbf062a251ed798391dd5099.jpg</td>\n",
       "      <td>20</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8619</th>\n",
       "      <td>cb039abc1a4a814d074febda7fc72f88.jpg</td>\n",
       "      <td>20</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10373</th>\n",
       "      <td>6de210cb997d6d05c73bbf59b22d54ff.jpg</td>\n",
       "      <td>20</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10583</th>\n",
       "      <td>133bf282cd606e8866df945ce86ef3f9.jpg</td>\n",
       "      <td>20</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11630</th>\n",
       "      <td>59b3b2b85afb42bb20d663f9dde07181.jpg</td>\n",
       "      <td>20</td>\n",
       "      <td>41</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                   filename category_27 category\n",
       "1952   a3ebc60c3f99cbdb957c0e66ce542661.jpg          20       41\n",
       "3134   5b62c3aed0dcbfb0fea50041e9a232c8.jpg          20       41\n",
       "3791   4f4ac25f31a06933a6d10f127b3a7778.jpg          20       41\n",
       "6162   a6ffb9e99b156aed038f4023c458a279.jpg          20       41\n",
       "8019   d8d36c64fbf062a251ed798391dd5099.jpg          20       41\n",
       "8619   cb039abc1a4a814d074febda7fc72f88.jpg          20       41\n",
       "10373  6de210cb997d6d05c73bbf59b22d54ff.jpg          20       41\n",
       "10583  133bf282cd606e8866df945ce86ef3f9.jpg          20       41\n",
       "11630  59b3b2b85afb42bb20d663f9dde07181.jpg          20       41"
      ]
     },
     "execution_count": 84,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd_query = pd_submission_prev[(pd_submission_prev['category_{}'.format(prev)] == '20') & (pd_submission_prev['category'] == '41')]\n",
    "print(len(pd_query), len(pd_query)/len(pd_submission_prev))\n",
    "pd_query.head(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "30 0.002461841457410143\n"
     ]
    },
    {
     "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>filename</th>\n",
       "      <th>category_25</th>\n",
       "      <th>category</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>419</th>\n",
       "      <td>41b8d6c7b78a01f77ed909c27c081794.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1423</th>\n",
       "      <td>f95d477b242a327927d9ff5f8a72a9da.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1598</th>\n",
       "      <td>0273cedf12ddeab3fdef5bb2dde65955.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1944</th>\n",
       "      <td>89c9176f9bb746a60fb1d10a566c45b3.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1952</th>\n",
       "      <td>a3ebc60c3f99cbdb957c0e66ce542661.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2203</th>\n",
       "      <td>d881412afc4e613f8b28be4de46be149.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2336</th>\n",
       "      <td>d50642fbb88ddd98719e0b3b5eaa6cdd.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3004</th>\n",
       "      <td>c0ec76fd0d7bc7d2677481ea50b4e91f.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3134</th>\n",
       "      <td>5b62c3aed0dcbfb0fea50041e9a232c8.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3153</th>\n",
       "      <td>9e506bb1b5b8d2dc3348c92029df575e.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3304</th>\n",
       "      <td>421f9cb07d0f3c1238034d23a5070e91.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3872</th>\n",
       "      <td>a24f66e632cb59afc38fb90518865199.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4245</th>\n",
       "      <td>dd4a68c1aecad1dd97f673394b9cb152.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4446</th>\n",
       "      <td>035ceb00c4acd20613abb9fe49ee8e33.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6378</th>\n",
       "      <td>7ee995f0ae4489b20d908d601a6a474a.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6847</th>\n",
       "      <td>c62e079b6a7ff0ec03d59944ed79bd15.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6854</th>\n",
       "      <td>243254d0b8e41fe6d2a8b928df15fe6f.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7081</th>\n",
       "      <td>085c844b9c4c0030a8a1fa35d98eb45e.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8189</th>\n",
       "      <td>cee68966f5710c07013aa028d021533d.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8567</th>\n",
       "      <td>1ba47fad280dcfd5dcb9604b515492e7.jpg</td>\n",
       "      <td>41</td>\n",
       "      <td>20</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                  filename category_25 category\n",
       "419   41b8d6c7b78a01f77ed909c27c081794.jpg          41       20\n",
       "1423  f95d477b242a327927d9ff5f8a72a9da.jpg          41       20\n",
       "1598  0273cedf12ddeab3fdef5bb2dde65955.jpg          41       20\n",
       "1944  89c9176f9bb746a60fb1d10a566c45b3.jpg          41       20\n",
       "1952  a3ebc60c3f99cbdb957c0e66ce542661.jpg          41       20\n",
       "2203  d881412afc4e613f8b28be4de46be149.jpg          41       20\n",
       "2336  d50642fbb88ddd98719e0b3b5eaa6cdd.jpg          41       20\n",
       "3004  c0ec76fd0d7bc7d2677481ea50b4e91f.jpg          41       20\n",
       "3134  5b62c3aed0dcbfb0fea50041e9a232c8.jpg          41       20\n",
       "3153  9e506bb1b5b8d2dc3348c92029df575e.jpg          41       20\n",
       "3304  421f9cb07d0f3c1238034d23a5070e91.jpg          41       20\n",
       "3872  a24f66e632cb59afc38fb90518865199.jpg          41       20\n",
       "4245  dd4a68c1aecad1dd97f673394b9cb152.jpg          41       20\n",
       "4446  035ceb00c4acd20613abb9fe49ee8e33.jpg          41       20\n",
       "6378  7ee995f0ae4489b20d908d601a6a474a.jpg          41       20\n",
       "6847  c62e079b6a7ff0ec03d59944ed79bd15.jpg          41       20\n",
       "6854  243254d0b8e41fe6d2a8b928df15fe6f.jpg          41       20\n",
       "7081  085c844b9c4c0030a8a1fa35d98eb45e.jpg          41       20\n",
       "8189  cee68966f5710c07013aa028d021533d.jpg          41       20\n",
       "8567  1ba47fad280dcfd5dcb9604b515492e7.jpg          41       20"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd_query = pd_iba[(pd_iba['category_{}'.format(prev)] == '41') & (pd_iba['category'] == '20')]\n",
    "print(len(pd_query), len(pd_query)/len(pd_submission_prev))\n",
    "pd_query.head(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x648 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(16, 9))\n",
    "plt.hist(pd_submission_prev['category'].sort_values(), bins=np.arange(-1, 43)-0.5)\n",
    "plt.xlim(-1, 43)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x648 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(16, 9))\n",
    "plt.hist(pd_iba['category'].sort_values(), bins=np.arange(-1, 43)-0.5)\n",
    "plt.xlim(-1, 43)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.8"
  },
  "toc-showmarkdowntxt": false
 },
 "nbformat": 4,
 "nbformat_minor": 4
}