homework8_20211122.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "11842944",
   "metadata": {},
   "source": [
    "## Dataset\n",
    "\n",
    "In this homework, we'll build a model for predicting if we have an image of a dog or a cat. For this, we will use the \"Dogs & Cats\" dataset that can be downloaded from Kaggle.\n",
    "\n",
    "You need to download the train.zip file.\n",
    "\n",
    "If you have troubles downloading from Kaggle, use this link instead:\n",
    "\n",
    "wget https://github.com/alexeygrigorev/large-datasets/releases/download/dogs-cats/train.zip\n",
    "In the lectures we saw how to use a pre-trained neural network. In the homework, we'll train a much smaller model from scratch.\n",
    "\n",
    "Note: You don't need a computer with a GPU for this homework. A laptop or any personal computer should be sufficient.\n",
    "\n",
    "## Data Preparation\n",
    "\n",
    "The dataset contains 12,500 images of cats and 12,500 images of dogs.\n",
    "\n",
    "Now we need to split this data into train and validation\n",
    "\n",
    "Create a train and validation folders\n",
    "In each folder, create cats and dogs folders\n",
    "Move the first 10,000 images to the train folder (from 0 to 9999) for boths cats and dogs - and put them in respective folders\n",
    "Move the remaining 2,500 images to the validation folder (from 10000 to 12499)\n",
    "You can do this manually or with Python (check os and shutil packages).\n",
    "\n",
    "## Model\n",
    "\n",
    "For this homework we will use Convolutional Neural Network (CNN. Like in the lectures, we'll use Keras.\n",
    "\n",
    "You need to develop the model with following structure:\n",
    "\n",
    "The shape for input should be (150, 150, 3)\n",
    "Next, create a covolutional layer (Conv2D):\n",
    "Use 32 filters\n",
    "Kernel size should be (3, 3) (that's the size of the filter)\n",
    "Use 'relu' as activation\n",
    "Reduce the size of the feature map with max pooling (MaxPooling2D)\n",
    "Set the pooling size to (2, 2)\n",
    "Turn the multi-dimensional result into vectors using a Flatten layer\n",
    "Next, add a Dense layer with 64 neurons and 'relu' activation\n",
    "Finally, create the Dense layer with 1 neuron - this will be the output\n",
    "The output layer should have an activation - use the appropriate activation for the binary classification case\n",
    "As optimizer use SGD with the following parameters:\n",
    "\n",
    "SGD(lr=0.002, momentum=0.8)\n",
    "For clarification about kernel size and max pooling, check Week #11 Office Hours.\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "6e2f87cb",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "2e5222c7",
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf\n",
    "from tensorflow import keras"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "46a678fe",
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.preprocessing.image import load_img\n",
    "from tensorflow.keras.applications.xception import Xception\n",
    "from tensorflow.keras.applications.xception import preprocess_input\n",
    "from tensorflow.keras.applications.xception import decode_predictions\n",
    "from tensorflow.keras.preprocessing.image import ImageDataGenerator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "0eb54a15",
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Dense, Flatten, Conv2D, MaxPooling2D\n",
    "from tensorflow.keras.losses import sparse_categorical_crossentropy\n",
    "from tensorflow.keras.optimizers import SGD"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ffa872bb",
   "metadata": {},
   "source": [
    "## Question 1\n",
    "\n",
    "Since we have a binary classification problem, what is the best loss function for us?\n",
    "\n",
    "Note: since we specify an activation for the output layer, we don't need to set from_logits=True\n",
    "\n",
    "Answer: BinaryCrossEntropy"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "30267531",
   "metadata": {},
   "source": [
    "## Question 2\n",
    "\n",
    "What's the total number of parameters of the model? You can use the summary method for that.\n",
    "\n",
    "Answer: 11,215,873"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "7485485e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "_________________________________________________________________\n",
      " Layer (type)                Output Shape              Param #   \n",
      "=================================================================\n",
      " input_1 (InputLayer)        [(None, 150, 150, 3)]     0         \n",
      "                                                                 \n",
      " conv_1 (Conv2D)             (None, 148, 148, 32)      896       \n",
      "                                                                 \n",
      " max_1 (MaxPooling2D)        (None, 74, 74, 32)        0         \n",
      "                                                                 \n",
      " flatten (Flatten)           (None, 175232)            0         \n",
      "                                                                 \n",
      " dense_1 (Dense)             (None, 64)                11214912  \n",
      "                                                                 \n",
      " output (Dense)              (None, 1)                 65        \n",
      "                                                                 \n",
      "=================================================================\n",
      "Total params: 11,215,873\n",
      "Trainable params: 11,215,873\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "def create_model():\n",
    "    inputs = keras.Input(shape=(150,150,3))\n",
    "\n",
    "    x = Conv2D(32, kernel_size=(3,3), activation='relu', name='conv_1')(inputs)\n",
    "    x = MaxPooling2D(pool_size=(2,2), name='max_1')(x)\n",
    "    flatten = Flatten(name='flatten')(x)\n",
    "    x = Dense(64, activation='relu', name='dense_1')(flatten)\n",
    "\n",
    "    outputs = Dense(1, activation='sigmoid', name='output')(x)\n",
    "    model = keras.Model(inputs, outputs)\n",
    "    return model\n",
    "\n",
    "model = create_model()\n",
    "model.summary()\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eff675ee",
   "metadata": {},
   "source": [
    "## Generators and Training\n",
    "\n",
    "For the next two questions, use the following data generator for both train and validation:\n",
    "\n",
    "ImageDataGenerator(rescale=1./255)\n",
    "We don't need to do any additional pre-processing for the images.\n",
    "When reading the data from train/val directories, check the class_mode parameter. Which value should it be for a binary classification problem?\n",
    "Use batch_size=20\n",
    "Use shuffle=True for both training and validaition\n",
    "For training use .fit() with the following params:\n",
    "\n",
    "model.fit(\n",
    "    train_generator,\n",
    "    steps_per_epoch=100,\n",
    "    epochs=10,\n",
    "    validation_data=validation_generator,\n",
    "    validation_steps=50\n",
    ")\n",
    "Note validation_steps=50 - this parameter says \"run only 50 steps on the validation data for evaluating the results\". This way we iterate a bit faster, but don't use the entire validation dataset. That's why it's important to shuffle the validation dataset as well."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "da5b2847",
   "metadata": {},
   "outputs": [],
   "source": [
    "opt = SGD(learning_rate=0.002, momentum=0.8)\n",
    "loss = keras.losses.BinaryCrossentropy()\n",
    "\n",
    "# Compile the model\n",
    "model.compile(optimizer=opt,\n",
    "              loss=loss,\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "ed6fbec8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 19995 images belonging to 2 classes.\n"
     ]
    }
   ],
   "source": [
    "train_gen = ImageDataGenerator(rescale=1./255)\n",
    "\n",
    "train_ds = train_gen.flow_from_directory(\n",
    "    '/Users/tonygunawan/Downloads/train/',\n",
    "    class_mode = 'binary',\n",
    "    target_size = (150, 150),\n",
    "    batch_size = 20,\n",
    "    shuffle = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "79c8d728",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'cats': 0, 'dogs': 1}"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_ds.class_indices"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "d54f4fd2",
   "metadata": {},
   "outputs": [],
   "source": [
    "X, y = next(train_ds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "fd073de5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[[[0.62352943, 0.7019608 , 0.65882355],\n",
       "         [0.61960787, 0.69803923, 0.6509804 ],\n",
       "         [0.6431373 , 0.70980394, 0.64705884],\n",
       "         ...,\n",
       "         [0.23137257, 0.27450982, 0.25882354],\n",
       "         [0.26666668, 0.3254902 , 0.2509804 ],\n",
       "         [0.3137255 , 0.38431376, 0.28235295]],\n",
       "\n",
       "        [[0.60784316, 0.6901961 , 0.6313726 ],\n",
       "         [0.6117647 , 0.69411767, 0.63529414],\n",
       "         [0.6431373 , 0.70980394, 0.64705884],\n",
       "         ...,\n",
       "         [0.21960786, 0.2627451 , 0.24705884],\n",
       "         [0.26666668, 0.3254902 , 0.2509804 ],\n",
       "         [0.31764707, 0.37647063, 0.29411766]],\n",
       "\n",
       "        [[0.6039216 , 0.6901961 , 0.6       ],\n",
       "         [0.6156863 , 0.7019608 , 0.61960787],\n",
       "         [0.64705884, 0.7137255 , 0.6509804 ],\n",
       "         ...,\n",
       "         [0.2392157 , 0.28235295, 0.26666668],\n",
       "         [0.3019608 , 0.35686275, 0.2901961 ],\n",
       "         [0.30980393, 0.3647059 , 0.30588236]],\n",
       "\n",
       "        ...,\n",
       "\n",
       "        [[0.15294118, 0.16078432, 0.11764707],\n",
       "         [0.20000002, 0.14901961, 0.11764707],\n",
       "         [0.14509805, 0.13725491, 0.08627451],\n",
       "         ...,\n",
       "         [0.2509804 , 0.27058825, 0.18431373],\n",
       "         [0.23529413, 0.2392157 , 0.25490198],\n",
       "         [0.2784314 , 0.28627452, 0.26666668]],\n",
       "\n",
       "        [[0.1254902 , 0.1137255 , 0.09411766],\n",
       "         [0.21568629, 0.13725491, 0.14117648],\n",
       "         [0.14901961, 0.12156864, 0.09019608],\n",
       "         ...,\n",
       "         [0.21960786, 0.2392157 , 0.15294118],\n",
       "         [0.24705884, 0.2509804 , 0.26666668],\n",
       "         [0.25882354, 0.26666668, 0.24705884]],\n",
       "\n",
       "        [[0.18431373, 0.14117648, 0.15686275],\n",
       "         [0.20000002, 0.10196079, 0.1254902 ],\n",
       "         [0.19215688, 0.13725491, 0.13725491],\n",
       "         ...,\n",
       "         [0.2627451 , 0.28235295, 0.19607845],\n",
       "         [0.23137257, 0.23529413, 0.2509804 ],\n",
       "         [0.2392157 , 0.2509804 , 0.22352943]]],\n",
       "\n",
       "\n",
       "       [[[0.77647066, 0.7960785 , 0.7803922 ],\n",
       "         [0.7725491 , 0.79215693, 0.77647066],\n",
       "         [0.75294125, 0.7725491 , 0.7568628 ],\n",
       "         ...,\n",
       "         [0.7294118 , 0.73333335, 0.7137255 ],\n",
       "         [0.7294118 , 0.73333335, 0.7137255 ],\n",
       "         [0.7176471 , 0.72156864, 0.7019608 ]],\n",
       "\n",
       "        [[0.7568628 , 0.77647066, 0.7607844 ],\n",
       "         [0.7607844 , 0.7803922 , 0.76470596],\n",
       "         [0.7607844 , 0.7803922 , 0.76470596],\n",
       "         ...,\n",
       "         [0.74509805, 0.7490196 , 0.7294118 ],\n",
       "         [0.7686275 , 0.7725491 , 0.75294125],\n",
       "         [0.7725491 , 0.77647066, 0.7568628 ]],\n",
       "\n",
       "        [[0.7686275 , 0.78823537, 0.7725491 ],\n",
       "         [0.7725491 , 0.79215693, 0.77647066],\n",
       "         [0.7725491 , 0.79215693, 0.77647066],\n",
       "         ...,\n",
       "         [0.78823537, 0.79215693, 0.7725491 ],\n",
       "         [0.7568628 , 0.7607844 , 0.7411765 ],\n",
       "         [0.73333335, 0.7372549 , 0.7176471 ]],\n",
       "\n",
       "        ...,\n",
       "\n",
       "        [[0.09803922, 0.09803922, 0.09019608],\n",
       "         [0.09411766, 0.09411766, 0.08627451],\n",
       "         [0.1137255 , 0.1137255 , 0.10588236],\n",
       "         ...,\n",
       "         [0.69411767, 0.67058825, 0.6784314 ],\n",
       "         [0.6901961 , 0.6666667 , 0.6745098 ],\n",
       "         [0.6784314 , 0.654902  , 0.654902  ]],\n",
       "\n",
       "        [[0.11764707, 0.11764707, 0.10980393],\n",
       "         [0.10196079, 0.10196079, 0.09411766],\n",
       "         [0.09803922, 0.09803922, 0.09019608],\n",
       "         ...,\n",
       "         [0.64705884, 0.62352943, 0.6313726 ],\n",
       "         [0.654902  , 0.6313726 , 0.6392157 ],\n",
       "         [0.67058825, 0.64705884, 0.64705884]],\n",
       "\n",
       "        [[0.1137255 , 0.1137255 , 0.10588236],\n",
       "         [0.09803922, 0.09803922, 0.09019608],\n",
       "         [0.10980393, 0.10980393, 0.10196079],\n",
       "         ...,\n",
       "         [0.62352943, 0.6       , 0.60784316],\n",
       "         [0.62352943, 0.6       , 0.60784316],\n",
       "         [0.6156863 , 0.5921569 , 0.5921569 ]]],\n",
       "\n",
       "\n",
       "       [[[0.5411765 , 0.6431373 , 0.6313726 ],\n",
       "         [0.6431373 , 0.74509805, 0.73333335],\n",
       "         [0.627451  , 0.7294118 , 0.7176471 ],\n",
       "         ...,\n",
       "         [0.8078432 , 0.9568628 , 0.9686275 ],\n",
       "         [0.74509805, 0.8941177 , 0.9058824 ],\n",
       "         [0.77647066, 0.93725497, 0.9450981 ]],\n",
       "\n",
       "        [[0.627451  , 0.7254902 , 0.7137255 ],\n",
       "         [0.63529414, 0.73333335, 0.72156864],\n",
       "         [0.6745098 , 0.7725491 , 0.7607844 ],\n",
       "         ...,\n",
       "         [0.81568635, 0.9725491 , 0.9803922 ],\n",
       "         [0.77647066, 0.9333334 , 0.94117653],\n",
       "         [0.8000001 , 0.9607844 , 0.9686275 ]],\n",
       "\n",
       "        [[0.6039216 , 0.6745098 , 0.6745098 ],\n",
       "         [0.6117647 , 0.68235296, 0.68235296],\n",
       "         [0.6431373 , 0.7137255 , 0.7137255 ],\n",
       "         ...,\n",
       "         [0.75294125, 0.91372555, 0.9215687 ],\n",
       "         [0.7294118 , 0.89019614, 0.8980393 ],\n",
       "         [0.77647066, 0.9490197 , 0.95294124]],\n",
       "\n",
       "        ...,\n",
       "\n",
       "        [[0.08627451, 0.05490196, 0.04705883],\n",
       "         [0.07058824, 0.05490196, 0.04313726],\n",
       "         [0.10196079, 0.10588236, 0.08627451],\n",
       "         ...,\n",
       "         [0.8313726 , 0.97647065, 0.9490197 ],\n",
       "         [0.7411765 , 0.87843144, 0.854902  ],\n",
       "         [0.75294125, 0.89019614, 0.86666673]],\n",
       "\n",
       "        [[0.08235294, 0.0509804 , 0.04313726],\n",
       "         [0.08627451, 0.07058824, 0.05882353],\n",
       "         [0.05490196, 0.05882353, 0.03921569],\n",
       "         ...,\n",
       "         [0.7490196 , 0.8941177 , 0.86666673],\n",
       "         [0.79215693, 0.9294118 , 0.9058824 ],\n",
       "         [0.75294125, 0.89019614, 0.86666673]],\n",
       "\n",
       "        [[0.0509804 , 0.01960784, 0.01176471],\n",
       "         [0.07058824, 0.05490196, 0.04313726],\n",
       "         [0.04705883, 0.0509804 , 0.03137255],\n",
       "         ...,\n",
       "         [0.7568628 , 0.90196085, 0.8745099 ],\n",
       "         [0.74509805, 0.882353  , 0.8588236 ],\n",
       "         [0.72156864, 0.8588236 , 0.8352942 ]]],\n",
       "\n",
       "\n",
       "       ...,\n",
       "\n",
       "\n",
       "       [[[0.04313726, 0.06666667, 0.20784315],\n",
       "         [0.15294118, 0.1764706 , 0.3019608 ],\n",
       "         [0.15686275, 0.18431373, 0.29803923],\n",
       "         ...,\n",
       "         [0.56078434, 0.5647059 , 0.6862745 ],\n",
       "         [0.56078434, 0.57254905, 0.6313726 ],\n",
       "         [0.21176472, 0.27450982, 0.33333334]],\n",
       "\n",
       "        [[0.0509804 , 0.07450981, 0.20000002],\n",
       "         [0.08235294, 0.10980393, 0.21960786],\n",
       "         [0.11764707, 0.14509805, 0.24705884],\n",
       "         ...,\n",
       "         [0.60784316, 0.6431373 , 0.7725491 ],\n",
       "         [0.40784317, 0.4431373 , 0.50980395],\n",
       "         [0.24313727, 0.30588236, 0.4039216 ]],\n",
       "\n",
       "        [[0.04705883, 0.07843138, 0.16862746],\n",
       "         [0.0509804 , 0.08235294, 0.16470589],\n",
       "         [0.07450981, 0.10588236, 0.18823531],\n",
       "         ...,\n",
       "         [0.48235297, 0.5058824 , 0.6       ],\n",
       "         [0.6392157 , 0.6627451 , 0.70980394],\n",
       "         [0.20784315, 0.27450982, 0.37647063]],\n",
       "\n",
       "        ...,\n",
       "\n",
       "        [[0.7058824 , 0.62352943, 0.4156863 ],\n",
       "         [0.6313726 , 0.5254902 , 0.41176474],\n",
       "         [0.70980394, 0.6039216 , 0.56078434],\n",
       "         ...,\n",
       "         [0.3647059 , 0.3921569 , 0.42352945],\n",
       "         [0.3372549 , 0.3647059 , 0.38823533],\n",
       "         [0.3372549 , 0.31764707, 0.39607847]],\n",
       "\n",
       "        [[0.59607846, 0.5372549 , 0.46274513],\n",
       "         [0.57254905, 0.5019608 , 0.44705886],\n",
       "         [0.6039216 , 0.54901963, 0.5058824 ],\n",
       "         ...,\n",
       "         [0.3254902 , 0.32156864, 0.3529412 ],\n",
       "         [0.37254903, 0.3647059 , 0.4156863 ],\n",
       "         [0.38823533, 0.40000004, 0.43529415]],\n",
       "\n",
       "        [[0.63529414, 0.6       , 0.5647059 ],\n",
       "         [0.5019608 , 0.45098042, 0.427451  ],\n",
       "         [0.45882356, 0.41960788, 0.4156863 ],\n",
       "         ...,\n",
       "         [0.3372549 , 0.33333334, 0.35686275],\n",
       "         [0.2627451 , 0.25490198, 0.30588236],\n",
       "         [0.49803925, 0.50980395, 0.54509807]]],\n",
       "\n",
       "\n",
       "       [[[0.43137258, 0.28627452, 0.17254902],\n",
       "         [0.46274513, 0.3254902 , 0.20784315],\n",
       "         [0.33333334, 0.20392159, 0.08235294],\n",
       "         ...,\n",
       "         [0.43529415, 0.4431373 , 0.5254902 ],\n",
       "         [0.37254903, 0.38431376, 0.4431373 ],\n",
       "         [0.45882356, 0.47450984, 0.52156866]],\n",
       "\n",
       "        [[0.43529415, 0.2901961 , 0.1764706 ],\n",
       "         [0.45098042, 0.3137255 , 0.19607845],\n",
       "         [0.33333334, 0.20392159, 0.08235294],\n",
       "         ...,\n",
       "         [0.58431375, 0.59607846, 0.6627451 ],\n",
       "         [0.28235295, 0.29803923, 0.34509805],\n",
       "         [0.0509804 , 0.06666667, 0.10196079]],\n",
       "\n",
       "        [[0.44705886, 0.3019608 , 0.18823531],\n",
       "         [0.44705886, 0.30980393, 0.19215688],\n",
       "         [0.33333334, 0.20392159, 0.08235294],\n",
       "         ...,\n",
       "         [0.16078432, 0.16862746, 0.21568629],\n",
       "         [0.08627451, 0.09803922, 0.1254902 ],\n",
       "         [0.05882353, 0.07058824, 0.09019608]],\n",
       "\n",
       "        ...,\n",
       "\n",
       "        [[0.427451  , 0.4039216 , 0.41176474],\n",
       "         [0.38431376, 0.36078432, 0.36862746],\n",
       "         [0.37254903, 0.34901962, 0.35686275],\n",
       "         ...,\n",
       "         [0.01960784, 0.01176471, 0.01568628],\n",
       "         [0.01176471, 0.00392157, 0.00784314],\n",
       "         [0.01176471, 0.00392157, 0.00784314]],\n",
       "\n",
       "        [[0.3803922 , 0.35686275, 0.3647059 ],\n",
       "         [0.37647063, 0.3529412 , 0.36078432],\n",
       "         [0.3803922 , 0.35686275, 0.3647059 ],\n",
       "         ...,\n",
       "         [0.01568628, 0.00784314, 0.01176471],\n",
       "         [0.01960784, 0.01176471, 0.01568628],\n",
       "         [0.01960784, 0.01176471, 0.01568628]],\n",
       "\n",
       "        [[0.39607847, 0.37254903, 0.3803922 ],\n",
       "         [0.44705886, 0.42352945, 0.43137258],\n",
       "         [0.427451  , 0.4039216 , 0.41176474],\n",
       "         ...,\n",
       "         [0.01568628, 0.00784314, 0.01176471],\n",
       "         [0.01960784, 0.01176471, 0.01568628],\n",
       "         [0.01960784, 0.01176471, 0.01568628]]],\n",
       "\n",
       "\n",
       "       [[[0.5568628 , 0.50980395, 0.3137255 ],\n",
       "         [0.54509807, 0.49803925, 0.3019608 ],\n",
       "         [0.6039216 , 0.5568628 , 0.36078432],\n",
       "         ...,\n",
       "         [0.5686275 , 0.5176471 , 0.34509805],\n",
       "         [0.60784316, 0.5529412 , 0.40000004],\n",
       "         [0.57254905, 0.52156866, 0.34901962]],\n",
       "\n",
       "        [[0.56078434, 0.5137255 , 0.3254902 ],\n",
       "         [0.5647059 , 0.5176471 , 0.32941177],\n",
       "         [0.53333336, 0.48627454, 0.29803923],\n",
       "         ...,\n",
       "         [0.5882353 , 0.5372549 , 0.3647059 ],\n",
       "         [0.5647059 , 0.5137255 , 0.34901962],\n",
       "         [0.5764706 , 0.5294118 , 0.33333334]],\n",
       "\n",
       "        [[0.56078434, 0.5137255 , 0.3254902 ],\n",
       "         [0.5411765 , 0.49411768, 0.30588236],\n",
       "         [0.5764706 , 0.5294118 , 0.34117648],\n",
       "         ...,\n",
       "         [0.58431375, 0.53333336, 0.35686275],\n",
       "         [0.5764706 , 0.5254902 , 0.3529412 ],\n",
       "         [0.5921569 , 0.54901963, 0.3372549 ]],\n",
       "\n",
       "        ...,\n",
       "\n",
       "        [[0.5647059 , 0.5372549 , 0.427451  ],\n",
       "         [0.5647059 , 0.5372549 , 0.427451  ],\n",
       "         [0.5568628 , 0.5294118 , 0.41960788],\n",
       "         ...,\n",
       "         [0.39607847, 0.34509805, 0.2784314 ],\n",
       "         [0.39607847, 0.33333334, 0.27058825],\n",
       "         [0.4039216 , 0.3529412 , 0.28627452]],\n",
       "\n",
       "        [[0.5529412 , 0.5137255 , 0.40784317],\n",
       "         [0.56078434, 0.52156866, 0.4156863 ],\n",
       "         [0.5529412 , 0.5137255 , 0.40784317],\n",
       "         ...,\n",
       "         [0.6       , 0.5529412 , 0.4901961 ],\n",
       "         [0.5568628 , 0.49411768, 0.43529415],\n",
       "         [0.5019608 , 0.45882356, 0.38823533]],\n",
       "\n",
       "        [[0.5686275 , 0.52156866, 0.43529415],\n",
       "         [0.5803922 , 0.53333336, 0.44705886],\n",
       "         [0.5803922 , 0.53333336, 0.44705886],\n",
       "         ...,\n",
       "         [0.50980395, 0.47450984, 0.40784317],\n",
       "         [0.5372549 , 0.48627454, 0.42352945],\n",
       "         [0.5882353 , 0.56078434, 0.48627454]]]], dtype=float32)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "0bebd10e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(20, 150, 150, 3)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "6038c751",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 5000 images belonging to 2 classes.\n"
     ]
    }
   ],
   "source": [
    "val_gen = ImageDataGenerator(rescale=1./255)\n",
    "\n",
    "val_ds = val_gen.flow_from_directory(\n",
    "    '/Users/tonygunawan/Downloads/validation/',\n",
    "    class_mode = 'binary',\n",
    "    target_size = (150, 150),\n",
    "    batch_size = 20,\n",
    "    shuffle = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "047472e1",
   "metadata": {},
   "outputs": [],
   "source": [
    "Y, z = next(val_ds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "8e56336c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(20, 150, 150, 3)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Y.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "601b6851",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "100/100 [==============================] - 29s 284ms/step - loss: 0.7009 - accuracy: 0.5123 - val_loss: 0.6962 - val_accuracy: 0.4930\n",
      "Epoch 2/10\n",
      "100/100 [==============================] - 28s 279ms/step - loss: 0.6927 - accuracy: 0.5090 - val_loss: 0.6934 - val_accuracy: 0.4730\n",
      "Epoch 3/10\n",
      "100/100 [==============================] - 30s 296ms/step - loss: 0.6922 - accuracy: 0.5165 - val_loss: 0.6914 - val_accuracy: 0.5750\n",
      "Epoch 4/10\n",
      "100/100 [==============================] - 32s 323ms/step - loss: 0.6899 - accuracy: 0.5535 - val_loss: 0.6865 - val_accuracy: 0.5600\n",
      "Epoch 5/10\n",
      "100/100 [==============================] - 31s 309ms/step - loss: 0.6810 - accuracy: 0.5555 - val_loss: 0.6819 - val_accuracy: 0.5490\n",
      "Epoch 6/10\n",
      "100/100 [==============================] - 31s 307ms/step - loss: 0.6755 - accuracy: 0.5645 - val_loss: 0.6839 - val_accuracy: 0.5440\n",
      "Epoch 7/10\n",
      "100/100 [==============================] - 31s 311ms/step - loss: 0.6709 - accuracy: 0.5935 - val_loss: 0.6664 - val_accuracy: 0.6050\n",
      "Epoch 8/10\n",
      "100/100 [==============================] - 33s 335ms/step - loss: 0.6664 - accuracy: 0.5955 - val_loss: 0.6572 - val_accuracy: 0.6080\n",
      "Epoch 9/10\n",
      "100/100 [==============================] - 34s 335ms/step - loss: 0.6551 - accuracy: 0.6225 - val_loss: 0.6502 - val_accuracy: 0.6100\n",
      "Epoch 10/10\n",
      "100/100 [==============================] - 31s 306ms/step - loss: 0.6420 - accuracy: 0.6245 - val_loss: 0.6652 - val_accuracy: 0.6030\n"
     ]
    }
   ],
   "source": [
    "history = model.fit(\n",
    "    train_ds,\n",
    "    steps_per_epoch=100,\n",
    "    epochs=10,\n",
    "    validation_data=val_ds,\n",
    "    validation_steps=50\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "552b5a55",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7fcdfc925910>"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "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": [
    "plt.plot(history.history['accuracy'], label='train')\n",
    "plt.plot(history.history['val_accuracy'], label='val')\n",
    "plt.xticks(np.arange(10))\n",
    "plt.legend()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d085bc8d",
   "metadata": {},
   "source": [
    "## Question 3\n",
    "\n",
    "What is the median of training accuracy for this model?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "679aac54",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.5599999725818634"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.median(history.history['accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d0e62fda",
   "metadata": {},
   "source": [
    "## Question 4\n",
    "\n",
    "What is the standard deviation of training loss for this model?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "e3743634",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.017573880369525282"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.std(history.history['loss'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fe70e606",
   "metadata": {},
   "source": [
    "## Data Augmentation\n",
    "\n",
    "For the next two questions, we'll generate more data using data augmentations.\n",
    "\n",
    "Add the following augmentations to your training data generator:\n",
    "\n",
    "rotation_range=40,\n",
    "width_shift_range=0.2,\n",
    "height_shift_range=0.2,\n",
    "shear_range=0.2,\n",
    "zoom_range=0.2,\n",
    "horizontal_flip=True,\n",
    "fill_mode='nearest'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "53d40db1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 19995 images belonging to 2 classes.\n"
     ]
    }
   ],
   "source": [
    "train_gen = ImageDataGenerator(\n",
    "    rescale=1./255,\n",
    "    rotation_range=40,\n",
    "    width_shift_range=0.2,\n",
    "    height_shift_range=0.2,\n",
    "    shear_range=0.2,\n",
    "    zoom_range=0.2,\n",
    "    horizontal_flip=True,\n",
    "    fill_mode='nearest'\n",
    ")\n",
    "\n",
    "train_ds = train_gen.flow_from_directory(\n",
    "    '/Users/tonygunawan/Downloads/train/',\n",
    "    class_mode = 'binary',\n",
    "    target_size = (150, 150),\n",
    "    batch_size = 20,\n",
    "    shuffle = True\n",
    "    \n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "45466792",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 5000 images belonging to 2 classes.\n"
     ]
    }
   ],
   "source": [
    "val_gen = ImageDataGenerator(rescale=1./255)\n",
    "\n",
    "val_ds = val_gen.flow_from_directory(\n",
    "    '/Users/tonygunawan/Downloads/validation/',\n",
    "    class_mode = 'binary',\n",
    "    target_size = (150, 150),\n",
    "    batch_size = 20,\n",
    "    shuffle = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "b0c45f8a",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "100/100 [==============================] - 50s 497ms/step - loss: 0.6751 - accuracy: 0.5810 - val_loss: 0.6785 - val_accuracy: 0.5600\n",
      "Epoch 2/10\n",
      "100/100 [==============================] - 48s 481ms/step - loss: 0.6653 - accuracy: 0.5890 - val_loss: 0.6495 - val_accuracy: 0.6150\n",
      "Epoch 3/10\n",
      "100/100 [==============================] - 46s 454ms/step - loss: 0.6682 - accuracy: 0.5870 - val_loss: 0.6587 - val_accuracy: 0.6010\n",
      "Epoch 4/10\n",
      "100/100 [==============================] - 42s 419ms/step - loss: 0.6583 - accuracy: 0.6080 - val_loss: 0.6292 - val_accuracy: 0.6410\n",
      "Epoch 5/10\n",
      "100/100 [==============================] - 45s 448ms/step - loss: 0.6683 - accuracy: 0.5785 - val_loss: 0.6630 - val_accuracy: 0.5770\n",
      "Epoch 6/10\n",
      "100/100 [==============================] - 40s 400ms/step - loss: 0.6533 - accuracy: 0.6070 - val_loss: 0.6300 - val_accuracy: 0.6460\n",
      "Epoch 7/10\n",
      "100/100 [==============================] - 49s 491ms/step - loss: 0.6624 - accuracy: 0.5875 - val_loss: 0.6311 - val_accuracy: 0.6310\n",
      "Epoch 8/10\n",
      "100/100 [==============================] - 44s 436ms/step - loss: 0.6507 - accuracy: 0.6160 - val_loss: 0.6257 - val_accuracy: 0.6600\n",
      "Epoch 9/10\n",
      "100/100 [==============================] - 45s 449ms/step - loss: 0.6559 - accuracy: 0.5965 - val_loss: 0.6310 - val_accuracy: 0.6470\n",
      "Epoch 10/10\n",
      "100/100 [==============================] - 45s 454ms/step - loss: 0.6527 - accuracy: 0.6025 - val_loss: 0.6541 - val_accuracy: 0.5940\n"
     ]
    }
   ],
   "source": [
    "history2 = model.fit(\n",
    "    train_ds,\n",
    "    steps_per_epoch=100,\n",
    "    epochs=10,\n",
    "    validation_data=val_ds,\n",
    "    validation_steps=50\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "7bfcb5a5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7fcdfc9252b0>"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "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": [
    "plt.plot(history2.history['accuracy'], label='train')\n",
    "plt.plot(history2.history['val_accuracy'], label='val')\n",
    "plt.xticks(np.arange(10))\n",
    "plt.legend()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f3826398",
   "metadata": {},
   "source": [
    "## Question 5\n",
    "\n",
    "Let's train our model for 10 more epochs using the same code as previously. Make sure you don't re-create the model - we want to continue training the model we already started training.\n",
    "\n",
    "What is the mean of validation loss for the model trained with augmentations?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "d8917e18",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6450831890106201"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.mean(history2.history['val_loss'])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "056d180d",
   "metadata": {},
   "source": [
    "## Question 6\n",
    "\n",
    "What's the average of validation accuracy for the last 5 epochs (from 6 to 10) for the model trained with augmentations?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "358b954c",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6416"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = [0.6690, 0.6700, 0.6130, 0.6350, 0.6210]\n",
    "\n",
    "np.average(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "17690d9d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6172000050544739"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.average(history2.history['val_accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bd77a5f8",
   "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": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}