{"cells": [{"outputs": [], "metadata": {"_uuid": "5038eff3a296cf0a73f40f6d2ce739b7082df35d", "_cell_guid": "764fe135-c09e-9769-5794-500867154d93", "_execution_state": "busy"}, "cell_type": "markdown", "source": "Starter EDA and ConvNet implementation using Keras. \n\nInspiration for this notebook comes from this [Keras blog post](https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html) and the [VGG ConvNet paper](https://arxiv.org/pdf/1409.1556.pdf). \n", "execution_count": null}, {"execution_count": null, "metadata": {"_uuid": "88b83f4c743c80f48280ac939e7a1ac2cfe7c8fa", "_cell_guid": "3d458c15-e131-f3c4-f756-843d6454bb37", "trusted": false}, "cell_type": "code", "source": "import os, cv2, random\nimport numpy as np\nimport pandas as pd\n\nimport matplotlib.pyplot as plt\nfrom matplotlib import ticker\nimport seaborn as sns\n%matplotlib inline \n\nfrom keras.models import Sequential\nfrom keras.layers import Input, Dropout, Flatten, Convolution2D, MaxPooling2D, Dense, Activation\nfrom keras.optimizers import RMSprop\nfrom keras.callbacks import ModelCheckpoint, Callback, EarlyStopping\nfrom keras.utils import np_utils", "outputs": []}, {"outputs": [], "metadata": {"_uuid": "f92b71b01d61696d472b510efa5084194a3e952e", "_cell_guid": "04681981-55ac-7820-a0ae-38f98c851c39"}, "cell_type": "markdown", "source": "## Preparing the Data\n\nThis function resizes the images to 64x64 and samples 2000 images (8%) of the data to run efficiently as a Kaggle Kernel. I also separated cats and dogs for exploratory analysis. ", "execution_count": null}, {"execution_count": null, "metadata": {"_uuid": "1f0eb2117663f0ba5c0002c00e4612987f1fc324", "_cell_guid": "663d335e-1b84-a8cb-19ee-8f04839cf4e5", "trusted": false}, "cell_type": "code", "source": "TRAIN_DIR = '../input/train/'\nTEST_DIR = '../input/test/'\n\nROWS = 64\nCOLS = 64\nCHANNELS = 3\n\ntrain_images = [TRAIN_DIR+i for i in os.listdir(TRAIN_DIR)] # use this for full dataset\ntrain_dogs =   [TRAIN_DIR+i for i in os.listdir(TRAIN_DIR) if 'dog' in i]\ntrain_cats =   [TRAIN_DIR+i for i in os.listdir(TRAIN_DIR) if 'cat' in i]\n\ntest_images =  [TEST_DIR+i for i in os.listdir(TEST_DIR)]\n\n\n# slice datasets for memory efficiency on Kaggle Kernels, delete if using full dataset\ntrain_images = train_dogs[:1000] + train_cats[:1000]\nrandom.shuffle(train_images)\ntest_images =  test_images[:25]\n\ndef read_image(file_path):\n    img = cv2.imread(file_path, cv2.IMREAD_COLOR) #cv2.IMREAD_GRAYSCALE\n    return cv2.resize(img, (ROWS, COLS), interpolation=cv2.INTER_CUBIC)\n\n\ndef prep_data(images):\n    count = len(images)\n    data = np.ndarray((count, CHANNELS, ROWS, COLS), dtype=np.uint8)\n\n    for i, image_file in enumerate(images):\n        image = read_image(image_file)\n        data[i] = image.T\n        if i%250 == 0: print('Processed {} of {}'.format(i, count))\n    \n    return data\n\ntrain = prep_data(train_images)\ntest = prep_data(test_images)\n\nprint(\"Train shape: {}\".format(train.shape))\nprint(\"Test shape: {}\".format(test.shape))", "outputs": []}, {"outputs": [], "metadata": {"_uuid": "e02473b88a7898d1015ec6a5784afa870bed7fab", "_cell_guid": "ed0fc95a-53bc-3517-245d-cf1f5122bc2d"}, "cell_type": "markdown", "source": "### Generating the Labels\n\nWe're dealing with a binary classification problem here - (1) dog (0) cat. The lables can be created by looping over the file names in the train directory. It's nice to see the training data is perfectly balanced. ", "execution_count": null}, {"execution_count": null, "metadata": {"_uuid": "56923aef75f4f8687bcfe58a4877419a9b113719", "_cell_guid": "c0f2fbf6-8d78-7d42-6579-5486a36c1e60", "trusted": false}, "cell_type": "code", "source": "labels = []\nfor i in train_images:\n    if 'dog' in i:\n        labels.append(1)\n    else:\n        labels.append(0)\n\nsns.countplot(labels)\nsns.plt.title('Cats and Dogs')", "outputs": []}, {"outputs": [], "metadata": {"_uuid": "8e600fdacf9e19f1e3fd7f058bed00c0f780bcc2", "_cell_guid": "4bb0bd2a-0512-aa72-c628-5b6ac946d97c"}, "cell_type": "markdown", "source": "### Checking out Cats and Dogs\nA quick side-by-side comparison of the animals.", "execution_count": null}, {"execution_count": null, "metadata": {"_uuid": "fbf37687abb052fe829267c4a5b1111e4965ff30", "_cell_guid": "0b7c79ae-6543-2ed8-e3f5-af4f5beb9cf7", "trusted": false}, "cell_type": "code", "source": "def show_cats_and_dogs(idx):\n    cat = read_image(train_cats[idx])\n    dog = read_image(train_dogs[idx])\n    pair = np.concatenate((cat, dog), axis=1)\n    plt.figure(figsize=(10,5))\n    plt.imshow(pair)\n    plt.show()\n    \nfor idx in range(0,5):\n    show_cats_and_dogs(idx)", "outputs": []}, {"outputs": [], "metadata": {"_uuid": "f0ef26b014765b1997a93243e73c99acfcd1d62a", "_cell_guid": "599a11ba-aaa4-8266-03ea-d816518e2a81"}, "cell_type": "markdown", "source": "### Your Average Cat and Dog Photo\n\nJust for fun, the mean pixel values for cats and dogs. I can almost see a resemblance, a ghost, if you will...", "execution_count": null}, {"execution_count": null, "metadata": {"_uuid": "d1312a5a112bc22a3df596ff4bf87e8bd4f80788", "_cell_guid": "7d3c6eb7-c370-9752-67c8-c780bab14abe", "trusted": false}, "cell_type": "code", "source": "dog_avg = np.array([dog[0].T for i, dog in enumerate(train) if labels[i]==1]).mean(axis=0)\nplt.imshow(dog_avg)\nplt.title('Your Average Dog')", "outputs": []}, {"execution_count": null, "metadata": {"_uuid": "51a4192525695c9aeb102e74d27da6756cad3865", "_cell_guid": "a5a93063-de92-18ce-b48f-68a193e90002", "trusted": false}, "cell_type": "code", "source": "cat_avg = np.array([cat[0].T for i, cat in enumerate(train) if labels[i]==0]).mean(axis=0)\nplt.imshow(cat_avg)\nplt.title('Your Average Cat')", "outputs": []}, {"outputs": [], "metadata": {"_uuid": "f5dd3e399da5984573f1f4d22ccd259a34516ca3", "_cell_guid": "51e403b6-bcfc-b4fa-3770-9850cc86bae3"}, "cell_type": "markdown", "source": "## CatdogNet-16\n\nA scaled down version of the VGG-16, with a few notable changes.\n\n- Number of convolution filters cut in half, fully connected (dense) layers scaled down. \n- Optimizer changed to `RMSprop`. \n- Output layer activation set to `sigmoid` for binary crossentropy. \n- Some layers commented out for efficiency.\n\nThe full network takes about 80s per epoch on a GTX1070 (or 2hr+ on CPU) on the full dataset.  (This script only trains on 8% of the 25K images. )", "execution_count": null}, {"execution_count": null, "metadata": {"_uuid": "c1328e689c76bd0f76bd257a2b7d8b275799bead", "_cell_guid": "4c477612-41d3-3112-5176-3c4ad9633080", "trusted": false}, "cell_type": "code", "source": "optimizer = RMSprop(lr=1e-4)\nobjective = 'binary_crossentropy'\nfrom keras import backend as K\nK.set_image_dim_ordering('th')\n\ndef catdog():\n    \n    model = Sequential()\n\n    model.add(Convolution2D(32, 3, 3, border_mode='same', input_shape=(3, ROWS, COLS), activation='relu'))\n    model.add(Convolution2D(32, 3, 3, border_mode='same', activation='relu'))\n    model.add(MaxPooling2D(pool_size=(2, 2)))\n\n    model.add(Convolution2D(64, 3, 3, border_mode='same', activation='relu'))\n    model.add(Convolution2D(64, 3, 3, border_mode='same', activation='relu'))\n    model.add(MaxPooling2D(pool_size=(2, 2)))\n    \n    model.add(Convolution2D(128, 3, 3, border_mode='same', activation='relu'))\n    model.add(Convolution2D(128, 3, 3, border_mode='same', activation='relu'))\n    model.add(MaxPooling2D(pool_size=(2, 2)))\n    \n    model.add(Convolution2D(256, 3, 3, border_mode='same', activation='relu'))\n    model.add(Convolution2D(256, 3, 3, border_mode='same', activation='relu'))\n    model.add(MaxPooling2D(pool_size=(2, 2)))\n\n    model.add(Flatten())\n    model.add(Dense(256, activation='relu'))\n    model.add(Dropout(0.5))\n    \n    model.add(Dense(256, activation='relu'))\n    model.add(Dropout(0.5))\n\n    model.add(Dense(1))\n    model.add(Activation('sigmoid'))\n\n    model.compile(loss=objective, optimizer=optimizer, metrics=['accuracy'])\n    return model\n\n\nmodel = catdog()", "outputs": []}, {"outputs": [], "metadata": {"_uuid": "37d11743f609b6c4fed35a0b47876dec2016cb3f", "_cell_guid": "b7fbf156-2268-62ce-5fed-52a09af5e6f7"}, "cell_type": "markdown", "source": "### Train and Predict\n\nI'm using Keras's early stopping callback to end training when the validation loss stops improving, otherwise the model will overfit. I will also be tracking the loss history on each epoch to visualize the overfitting trend. \n\nNote: A slice of 1000 images was used to fit the model for CPU efficency. The model's perfrmance improves significantly when used on the entire dataset. ", "execution_count": null}, {"execution_count": null, "metadata": {"_uuid": "eedcbb00c03ca614fd0f8b104f1a6baf17b6aa2d", "_cell_guid": "ca613169-4b41-e9d5-27c7-6629f8e035c8", "trusted": false}, "cell_type": "code", "source": "nb_epoch = 10\nbatch_size = 16\n\n## Callback for loss logging per epoch\nclass LossHistory(Callback):\n    def on_train_begin(self, logs={}):\n        self.losses = []\n        self.val_losses = []\n        \n    def on_epoch_end(self, batch, logs={}):\n        self.losses.append(logs.get('loss'))\n        self.val_losses.append(logs.get('val_loss'))\n\nearly_stopping = EarlyStopping(monitor='val_loss', patience=3, verbose=1, mode='auto')        \n        \ndef run_catdog():\n    \n    history = LossHistory()\n    model.fit(train, labels, batch_size=batch_size, nb_epoch=nb_epoch,\n              validation_split=0.25, verbose=0, shuffle=True, callbacks=[history, early_stopping])\n    \n\n    predictions = model.predict(test, verbose=0)\n    return predictions, history\n\npredictions, history = run_catdog()", "outputs": []}, {"execution_count": null, "metadata": {"collapsed": true, "_uuid": "4c51344e65a669c1213ea78f2c9e720c90fe7d04", "trusted": false, "_cell_guid": "6266d01c-9d4c-a1dc-8cb9-ce7b3107b536"}, "cell_type": "code", "source": "loss = history.losses\nval_loss = history.val_losses\n\nplt.xlabel('Epochs')\nplt.ylabel('Loss')\nplt.title('VGG-16 Loss Trend')\nplt.plot(loss, 'blue', label='Training Loss')\nplt.plot(val_loss, 'green', label='Validation Loss')\nplt.xticks(range(0,nb_epoch)[0::2])\nplt.legend()\nplt.show()", "outputs": []}, {"outputs": [], "metadata": {"_uuid": "800352efd3f2eb6db9f660c637e68fa0da73727d", "_cell_guid": "d9382e7d-31db-0d75-6876-aaccb126e4f6"}, "cell_type": "markdown", "source": "## How'd We Do?\n\nI'm pretty sure I can distinguish a cat from a dog 100% of the time, but how confident is the model?...\n\nTip: Run on the full dataset with a GPU for a LB logloss of ~0.4 and accuracy at approx 90%. ", "execution_count": null}, {"execution_count": null, "metadata": {"collapsed": true, "_uuid": "1812c012d16bd635f1d310a2813928de17ad6aba", "trusted": false, "_cell_guid": "c665d01b-a5a7-2f7e-e5d4-09e108920e40"}, "cell_type": "code", "source": "for i in range(0,10):\n    if predictions[i, 0] >= 0.5: \n        print('I am {:.2%} sure this is a Dog'.format(predictions[i][0]))\n    else: \n        print('I am {:.2%} sure this is a Cat'.format(1-predictions[i][0]))\n        \n    plt.imshow(test[i].T)\n    plt.show()", "outputs": []}], "nbformat": 4, "metadata": {"language_info": {"version": "3.6.1", "mimetype": "text/x-python", "file_extension": ".py", "nbconvert_exporter": "python", "codemirror_mode": {"version": 3, "name": "ipython"}, "pygments_lexer": "ipython3", "name": "python"}, "_change_revision": 0, "_is_fork": false, "kernelspec": {"display_name": "Python 3", "name": "python3", "language": "python"}}, "nbformat_minor": 1}