{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "## The CIFAR10 Dataset\n", "\n", "[Keras CIFAR10 info page](https://keras.io/api/datasets/cifar10)\n", "          \n", "[CIFAR10 home page](https://www.cs.toronto.edu/~kriz/cifar.html)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Preliminaries" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import random\n", "import numpy as np\n", "import matplotlib.pyplot as plt\n", "plt.rcParams[\"figure.figsize\"] = (2,2) # default figure size: 2x2 inches" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "kKKAa30Y8NvM", "jupyter": { "source_hidden": true }, "tags": [] }, "outputs": [], "source": [ "# generalized version of plot_history that plots validation data if available\n", "\n", "def plot_history(history):\n", " loss_values = history.history['loss']\n", " accuracy_values = history.history['accuracy']\n", " validation = 'val_loss' in history.history\n", " if validation:\n", " val_loss_values = history.history['val_loss']\n", " val_accuracy_values = history.history['val_accuracy']\n", " epoch_nums = range(1, len(loss_values)+1)\n", " plt.figure(figsize=(12,4)) # width, height in inches\n", " plt.subplot(1, 2, 1)\n", " if validation:\n", " plt.plot(epoch_nums, loss_values, 'r', label=\"Training loss\")\n", " plt.plot(epoch_nums, val_loss_values, 'r--', label=\"Validation loss\")\n", " plt.title(\"Training/validation loss\")\n", " plt.legend()\n", " else:\n", " plt.plot(epoch_nums, loss_values, 'r', label=\"Training loss\")\n", " plt.title(\"Training loss\")\n", " plt.xlabel(\"Epochs\")\n", " plt.ylabel(\"Loss\")\n", " plt.subplot(1, 2, 2)\n", " if validation:\n", " plt.plot(epoch_nums, accuracy_values, 'b', label='Training accuracy')\n", " plt.plot(epoch_nums, val_accuracy_values, 'b--', label='Validation accuracy')\n", " plt.title(\"Training/validation accuracy\")\n", " plt.legend()\n", " else:\n", " plt.plot(epoch_nums, accuracy_values, 'b', label='Training accuracy')\n", " plt.title(\"Training accuracy\")\n", " plt.xlabel(\"Epochs\")\n", " plt.ylabel(\"Accuracy\")\n", " plt.ylim(0, 1)\n", " plt.show()\n", " \n", "from tensorflow.keras.models import Model\n", "\n", "# this version of show_channels does not call preprocess_image\n", "\n", "def show_channels(network, image, layer_name, channels=range(20), cmap='gray', cols=5):\n", " # channels can be a number like 0 or a sequence like [0, 2, 4] or range(10)\n", " layer_names = [layer.name for layer in network.layers]\n", " if layer_name not in layer_names:\n", " print(f'No such layer: {layer_name}')\n", " return\n", " # generate activation maps for layer_name\n", " input_tensor = network.layers[0].input\n", " output_tensor = network.get_layer(layer_name).output\n", " activation_model = Model(inputs=input_tensor, outputs=output_tensor)\n", " batch = np.array([image])\n", " #output = activation_model.predict(batch)[0] # using predict causes a weird warning message\n", " output = activation_model(batch)[0].numpy()\n", " h, w, d = output.shape\n", " # display activation maps\n", " if type(channels) is int:\n", " channels = [channels]\n", " rows = len(channels) // cols\n", " if len(channels) > rows*cols:\n", " rows += 1\n", " plt.figure(figsize=(3*cols,3*rows)) # (width, height) in inches\n", " k = 0\n", " for channel in channels:\n", " if 0 <= channel < d:\n", " k += 1\n", " plt.subplot(rows, cols, k)\n", " plt.imshow(output[:,:,channel], cmap=cmap)\n", " plt.title(f'channel {channel}')\n", " plt.axis('off')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### CIFAR10 categories\n", "\n", "* 0 = airplane\n", "* 1 = car\n", "* 2 = bird\n", "* 3 = cat\n", "* 4 = deer\n", "* 5 = dog\n", "* 6 = frog\n", "* 7 = horse\n", "* 8 = ship\n", "* 9 = truck" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Load and examine the data" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from keras.datasets import cifar10" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "(train_images,train_labels), (test_images,test_labels) = cifar10.load_data()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "train_images.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "print(train_images.dtype, train_images.min(), train_images.max())" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "train_images[0]" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "plt.imshow(train_images[0]);" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "train_labels.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "print(train_labels[0])" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "train_labels[:10]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "print(train_labels[0][0])" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "train_labels = train_labels.reshape(50000)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "train_labels.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "print(train_labels[0])" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "train_labels[:10]" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "test_labels = test_labels.reshape(len(test_labels))" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "test_labels.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "names = 'airplane car bird cat deer dog frog horse ship truck'.split()" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "names" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "names[6]" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "jupyter": { "source_hidden": true } }, "outputs": [], "source": [ "def show_random_image():\n", " n = random.randrange(len(train_images))\n", " category_num = train_labels[n]\n", " category_name = names[category_num]\n", " print(f\"train_images[{n}]: {category_name}\")\n", " plt.imshow(train_images[n])" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "show_random_image()" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "8fVjEm9F8NvR", "jupyter": { "source_hidden": true } }, "outputs": [], "source": [ "# show_random_selection() shows training images from all categories\n", "# show_random_selection(n) shows training images from category n\n", "\n", "def show_random_selection(category_num=None):\n", " if category_num == None:\n", " which = range(len(train_images))\n", " elif 0 <= category_num < len(names):\n", " which = [i for i in range(len(train_images)) if train_labels[i] == category_num]\n", " else:\n", " print(\"category out of range\")\n", " return\n", " plt.figure(figsize=(15,15)) # (width, height) in inches\n", " rows, columns = 5, 6\n", " for k in range(1, columns*rows+1):\n", " i = random.choice(which)\n", " plt.subplot(rows, columns, k)\n", " category_num = train_labels[i]\n", " plt.title(names[category_num])\n", " plt.axis('off')\n", " plt.imshow(train_images[i])" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "show_random_selection(6)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "from keras.utils import to_categorical" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "# create the one-hot target vectors\n", "train_targets = to_categorical(train_labels, num_classes=10)\n", "test_targets = to_categorical(test_labels, num_classes=10)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "test_targets[99]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "print(test_labels[99])" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "names[7]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "plt.imshow(test_images[99]);" ] }, { "cell_type": "markdown", "metadata": { "jp-MarkdownHeadingCollapsed": true, "tags": [] }, "source": [ "### Optional: Convert images to grayscale" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "frog = train_images[0]" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "frog.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "plt.imshow(frog);" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "frog" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Let's start with just the top row of pixels in the image:" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "top_row = frog[0] # top row of pixels" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "top_row" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "len(top_row)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "We can average together the three RGB values for each pixel like this:" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "np.dot(top_row, [1/3, 1/3, 1/3])" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "len(np.dot(top_row, [1/3, 1/3, 1/3]))" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "np.round(np.dot(top_row, [1/3, 1/3, 1/3]))" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "np.round(np.dot(top_row, [1/3, 1/3, 1/3])).astype('uint8')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "We can also apply the operation to the whole image:" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "np.round(np.dot(frog, [1/3, 1/3, 1/3])).astype('uint8')" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "# convert a color image to grayscale\n", "\n", "def color2gray(image):\n", " assert image.dtype == 'uint8', \"image must be of type uint8\"\n", " rgb_weights = [1/3, 1/3, 1/3]\n", " gray = np.round(np.dot(image, rgb_weights)).astype('uint8')\n", " return gray" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "frog_gray = color2gray(frog)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "frog_gray.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "plt.imshow(frog_gray); # default colormap" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "plt.imshow(frog_gray, cmap='gray'); # grayscale colormap" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "# version 2:\n", "# - corrects for perceptual effects\n", "# - returns shape (height, width, 1)\n", "\n", "def color2gray(image):\n", " assert image.dtype == 'uint8', \"image must be of type uint8\"\n", " rgb_weights = [0.299, 0.587, 0.114] # corrects for perceptual effects\n", " gray = np.round(np.dot(image, rgb_weights)).astype('uint8')\n", " gray = gray.reshape(gray.shape + (1,)) # reshape as (height, width, 1)\n", " return gray" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "color2gray(frog).shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "plt.imshow(color2gray(frog), cmap='gray');" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "color2gray(train_images[0]).shape" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "We can convert all training images at once to grayscale:" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "color2gray(train_images).shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "train_images_gray = color2gray(train_images)\n", "test_images_gray = color2gray(test_images)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "test_images_gray.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "test_images_gray.dtype" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "plt.imshow(test_images_gray[99], cmap='gray');" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Prepare the training data" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "(train_images,train_labels), (test_images,test_labels) = cifar10.load_data()\n", "# simplify the labels\n", "train_labels = train_labels.reshape(len(train_labels))\n", "test_labels = test_labels.reshape(len(test_labels))" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "print(train_images.dtype, train_images.min(), train_images.max())" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "train_images.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "test_images.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "def normalize(image):\n", " assert image.dtype == 'uint8', \"image must be of type uint8\"\n", " return (image / 255).astype('float32')" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "normalize(train_images[0])" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "train_images = normalize(train_images)\n", "test_images = normalize(test_images)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "print(train_images.dtype, train_images.min(), train_images.max())" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "plt.imshow(train_images[0]);" ] }, { "cell_type": "markdown", "metadata": { "id": "oNye2dTw8NvS" }, "source": [ "### Build a convolutional neural network" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "1Y6w5f5a8NvS" }, "outputs": [], "source": [ "from keras.models import Sequential\n", "from keras.layers import Input, Dense, Flatten\n", "from keras.layers import Conv2D, MaxPooling2D" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "r3V1roX-8NvT", "jupyter": { "source_hidden": true } }, "outputs": [], "source": [ "def build_convnet():\n", " convnet = Sequential()\n", " convnet.add(Input(shape=(32,32,3)))\n", " convnet.add(Conv2D(64, (3,3), activation='relu', name='conv1'))\n", " convnet.add(MaxPooling2D((2,2), name='pool1'))\n", " convnet.add(Conv2D(64, (3,3), activation='relu', name='conv2'))\n", " convnet.add(MaxPooling2D((2,2), name='pool2'))\n", " convnet.add(Conv2D(64, (3,3), activation='relu', name='conv3'))\n", " convnet.add(Flatten())\n", " convnet.add(Dense(64, activation='relu', name='hidden'))\n", " convnet.add(Dense(10, activation='softmax', name='output'))\n", " \n", " convnet.compile(loss='categorical_crossentropy',\n", " optimizer='rmsprop',\n", " metrics=['accuracy'])\n", " return convnet" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "tZ4j2kQk8NvT" }, "outputs": [], "source": [ "convnet = build_convnet()" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "YI0bkp688NvT", "outputId": "df12f7f5-0e6c-49c9-ada3-d8a76371b70d" }, "outputs": [], "source": [ "convnet.summary()" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "QgPX6uDi8NvT", "outputId": "70203e3c-bf78-4e7f-d8d5-5bec9aee6a74", "tags": [] }, "outputs": [], "source": [ "history = convnet.fit(train_images, train_targets, epochs=10, batch_size=64)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "Ekhhebp28NvU", "outputId": "94648dc7-7515-4dde-d9fe-56b9729b5e33" }, "outputs": [], "source": [ "convnet.evaluate(train_images, train_targets)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "RENYg4MF8NvU", "outputId": "91c0895c-b9d1-4a55-ec5a-33a726542d8e" }, "outputs": [], "source": [ "convnet.evaluate(test_images, test_targets)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/", "height": 295 }, "id": "SQ-Rp79h8NvU", "outputId": "9dffca8d-fa9d-4241-ff7a-0bc59030aeeb" }, "outputs": [], "source": [ "plot_history(history)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "rIyT6ZhB8NvU" }, "outputs": [], "source": [ "outputs = convnet.predict(test_images)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "sNcqZR6p8NvU", "outputId": "6761222c-21f2-4114-ef73-3057f6e75275" }, "outputs": [], "source": [ "outputs.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "bMY5RDU28NvV" }, "outputs": [], "source": [ "predictions = [np.argmax(output) for output in outputs]" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "A6jEHSHR8NvV" }, "outputs": [], "source": [ "wrong = [i for i in range(len(predictions)) if predictions[i] != test_labels[i]]" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "Q3qYaTre8NvV", "outputId": "b7978012-2a4d-459c-91c4-1cee3fa0c87f" }, "outputs": [], "source": [ "print(f\"Misclassified {len(wrong)} test images out of {len(test_images)}\")" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "jupyter": { "source_hidden": true }, "tags": [] }, "outputs": [], "source": [ "def show_wrong_images(network):\n", " outputs = network.predict(test_images)\n", " predictions = [np.argmax(output) for output in outputs]\n", " wrong = [i for i in range(len(predictions)) if predictions[i] != test_labels[i]]\n", " print(f\"Misclassified {len(wrong)} test images out of {len(test_images)}\")\n", " plt.figure(figsize=(12,12)) # (width, height) in inches\n", " rows, columns = 5, 6\n", " for i in range(1, columns*rows+1):\n", " w = random.choice(wrong)\n", " img = test_images[w]\n", " predicted_name = names[predictions[w]]\n", " correct_name = names[test_labels[w]]\n", " plt.subplot(rows, columns, i)\n", " plt.title(f'\"{predicted_name}\" ({correct_name})')\n", " plt.axis('off')\n", " plt.imshow(img) #, cmap='gray')" ] }, { "cell_type": "markdown", "metadata": { "id": "xWzA_SzO8NvW", "tags": [] }, "source": [ "A random sampling of misclassified images:" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/", "height": 682 }, "id": "KcdUFpB28NvW", "outputId": "893a9785-78a0-4469-85be-bad65ad9d535" }, "outputs": [], "source": [ "show_wrong_images(convnet)" ] }, { "cell_type": "markdown", "metadata": { "id": "te3iyYJJB8vG" }, "source": [ "### Use a validation set" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "y-NxLKcHB_YF", "outputId": "e7733bbb-ec58-4897-c793-9c3e7cded645" }, "outputs": [], "source": [ "train_images.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "zZ8aIk0CCMzS", "outputId": "7f035fa9-9436-4388-f371-b4851f72b761" }, "outputs": [], "source": [ "train_labels.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "G1oSVXEGCCvv", "outputId": "1e8eff83-223f-4273-bb07-ef67bad1b345" }, "outputs": [], "source": [ "train_targets.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "rAPOb_rsCO4E" }, "outputs": [], "source": [ "# of images to use for validation set\n", "split = 10000\n", "\n", "val_images = train_images[:split]\n", "val_targets = train_targets[:split]\n", "val_labels = train_labels[:split]\n", "\n", "subset_images = train_images[split:]\n", "subset_targets = train_targets[split:]\n", "subset_labels = train_labels[split:]\n", "\n", "print(f\"Using {len(subset_images)} images for training, {len(val_images)} for validation\")" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "nFNA4PzYC4VF", "outputId": "61d52751-f82f-4e46-b096-0892985abd06" }, "outputs": [], "source": [ "subset_images.shape" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "WIblTME6Cxif" }, "outputs": [], "source": [ "convnet = build_convnet()" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "BEtb9bmnC63t", "outputId": "b030ed8e-df86-415f-b458-17180b611283", "tags": [] }, "outputs": [], "source": [ "history = convnet.fit(subset_images, subset_targets,\n", " validation_data=(val_images, val_targets),\n", " epochs=10, batch_size=64)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "fLPHDkw0DPeS", "outputId": "2aa5b1a8-8e0d-418d-bd60-7d209d7451d2" }, "outputs": [], "source": [ "convnet.evaluate(subset_images, subset_targets)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "bTI-mUNrDhMb", "outputId": "a5d334eb-0f7b-4875-9246-459542b0de1b" }, "outputs": [], "source": [ "convnet.evaluate(test_images, test_targets)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/", "height": 295 }, "id": "FYC4SsYRDm9Q", "outputId": "f7b4d6c7-54ac-4955-b52a-4bf49188ba7b" }, "outputs": [], "source": [ "plot_history(history)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "id": "JlrvfMqVDomg" }, "outputs": [], "source": [ "convnet = build_convnet()" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "E-iyF6QPDuJ3", "outputId": "8ead8451-26c0-4973-aca9-b9b896161e29" }, "outputs": [], "source": [ "history = convnet.fit(train_images, train_targets, epochs=4, batch_size=64)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "OZTKNdbPEdMV", "outputId": "412c7169-3b35-4b73-c2d8-2134ecc23913" }, "outputs": [], "source": [ "convnet.evaluate(train_images, train_targets)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "u2KL4xFZD1tO", "outputId": "c9efebac-cf90-43ce-eb6f-33abeac0f87a" }, "outputs": [], "source": [ "convnet.evaluate(test_images, test_targets)" ] }, { "cell_type": "markdown", "metadata": { "id": "wGHDd0wND9uN", "tags": [] }, "source": [ "### Cats vs. Dogs" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "jupyter": { "source_hidden": true } }, "outputs": [], "source": [ "# This function returns a subset of the CIFAR10 dataset\n", "# consisting of the specified image classes. \n", "\n", "def load_cifar10_subset(selected_classes=[]):\n", " if len(selected_classes) == 0:\n", " selected_classes = list(range(10))\n", " for n in selected_classes:\n", " assert type(n) == int and 0 <= n <= 9, \"classes must be ints in range 0-9\"\n", " # load the data\n", " (train_images,train_labels), (test_images,test_labels) = cifar10.load_data()\n", " # simplify the labels\n", " train_labels = train_labels.reshape(len(train_labels))\n", " test_labels = test_labels.reshape(len(test_labels))\n", " # extract the desired classes\n", " indices = [i for i in range(len(train_labels)) if train_labels[i] in selected_classes]\n", " selected_train_images = train_images[indices]\n", " selected_train_labels = train_labels[indices]\n", " indices = [i for i in range(len(test_labels)) if test_labels[i] in selected_classes]\n", " selected_test_images = test_images[indices]\n", " selected_test_labels = test_labels[indices]\n", " new_train_labels = np.array([selected_classes.index(i) for i in selected_train_labels])\n", " new_test_labels = np.array([selected_classes.index(i) for i in selected_test_labels])\n", " class_names = 'airplane car bird cat deer dog frog horse ship truck'.split()\n", " new_names = [class_names[n] for n in selected_classes]\n", " # normalize the images\n", " selected_train_images = normalize(selected_train_images)\n", " selected_test_images = normalize(selected_test_images)\n", " return ((selected_train_images,new_train_labels),\n", " (selected_test_images,new_test_labels),\n", " new_names)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "4wfox7ck8NvN", "outputId": "9ce0631e-029a-4959-b210-b64c11f9079d" }, "outputs": [], "source": [ "(train_images,train_labels), (test_images,test_labels), names = load_cifar10_subset([3,5])" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "train_labels[:20]" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "names" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "plt.imshow(train_images[5]);" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "plt.imshow(train_images[4]);" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "4wfox7ck8NvN", "outputId": "9ce0631e-029a-4959-b210-b64c11f9079d" }, "outputs": [], "source": [ "# create the one-hot target vectors\n", "train_targets = to_categorical(train_labels, num_classes=2)\n", "test_targets = to_categorical(test_labels, num_classes=2)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "train_targets[0]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "show_random_selection(0)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "show_random_selection(1)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "#### How well can you train the network to distinguish cats from dogs?\n", "\n", "Try it!" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "metadata": { "id": "wGHDd0wND9uN", "tags": [] }, "source": [ "### Dogs vs. Airplanes" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "4wfox7ck8NvN", "outputId": "9ce0631e-029a-4959-b210-b64c11f9079d", "tags": [] }, "outputs": [], "source": [ "(train_images,train_labels), (test_images,test_labels), names = load_cifar10_subset([5,0])\n", "\n", "# create the one-hot target vectors\n", "train_targets = to_categorical(train_labels, num_classes=2)\n", "test_targets = to_categorical(test_labels, num_classes=2)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "names" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "show_random_selection()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "#### How well can you train the network to distinguish dogs from airplanes?\n", "\n", "Try it!" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Cars vs. Trucks vs. Ships vs. Airplanes" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "4wfox7ck8NvN", "outputId": "9ce0631e-029a-4959-b210-b64c11f9079d", "tags": [] }, "outputs": [], "source": [ "(train_images,train_labels), (test_images,test_labels), names = load_cifar10_subset([1,9,8,0])\n", "\n", "# create the one-hot target vectors\n", "train_targets = to_categorical(train_labels, num_classes=4)\n", "test_targets = to_categorical(test_labels, num_classes=4)" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "tags": [] }, "outputs": [], "source": [ "show_random_selection()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Try it!" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Or some other combination of categories..." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "Python 3 (ipykernel)", "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": "3.12.12" } }, "nbformat": 4, "nbformat_minor": 4 }