{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"# Setup","metadata":{}},{"cell_type":"code","source":"import pandas as pd\nimport numpy as np\nimport matplotlib.pyplot as plt\nimport matplotlib.image as mpimg\nimport seaborn as sns\n%matplotlib inline\n\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import confusion_matrix\nimport itertools\n\nfrom keras.utils.np_utils import to_categorical\nfrom keras.models import Sequential\nfrom keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D, BatchNormalization\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom tensorflow import keras\nfrom tensorflow.keras import layers\nfrom keras.callbacks import ReduceLROnPlateau\n\nsns.set(style='white', context='notebook', palette='deep')","metadata":{"_uuid":"051d70d956493feee0c6d64651c6a088724dca2a","_execution_state":"idle","execution":{"iopub.status.busy":"2022-07-20T01:22:23.293535Z","iopub.execute_input":"2022-07-20T01:22:23.294005Z","iopub.status.idle":"2022-07-20T01:22:30.167184Z","shell.execute_reply.started":"2022-07-20T01:22:23.293934Z","shell.execute_reply":"2022-07-20T01:22:30.165719Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Pre-Processing","metadata":{}},{"cell_type":"code","source":"# Importing Data\ntrain = pd.read_csv(\"../input/digit-recognizer/train.csv\")\ntest = pd.read_csv(\"../input/digit-recognizer/test.csv\")\n\nY_train = train[\"label\"]\nX_train = train.drop(labels=[\"label\"], axis=1)\n\ng = sns.countplot(Y_train)\nY_train.value_counts()","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:30.169589Z","iopub.execute_input":"2022-07-20T01:22:30.170239Z","iopub.status.idle":"2022-07-20T01:22:35.738229Z","shell.execute_reply.started":"2022-07-20T01:22:30.170203Z","shell.execute_reply":"2022-07-20T01:22:35.737275Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"X_train.isnull().any().describe()\ntest.isnull().any().describe()","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:35.741328Z","iopub.execute_input":"2022-07-20T01:22:35.741655Z","iopub.status.idle":"2022-07-20T01:22:35.771106Z","shell.execute_reply.started":"2022-07-20T01:22:35.741628Z","shell.execute_reply":"2022-07-20T01:22:35.770169Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Normalize the data\nX_train = X_train / 255.0\ntest = test / 255.0","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:35.773656Z","iopub.execute_input":"2022-07-20T01:22:35.774591Z","iopub.status.idle":"2022-07-20T01:22:35.939534Z","shell.execute_reply.started":"2022-07-20T01:22:35.774560Z","shell.execute_reply":"2022-07-20T01:22:35.938506Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Reshape image in 3 dimensions (height = 28px, width = 28px , canal = 1)\nX_train = X_train.values.reshape(-1,28,28,1)\ntest = test.values.reshape(-1,28,28,1)\n\n# Encode labels to one hot vectors (ex : 2 -> [0,0,1,0,0,0,0,0,0,0])\nY_train = to_categorical(Y_train, num_classes=len(Y_train.value_counts()))","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:35.941226Z","iopub.execute_input":"2022-07-20T01:22:35.941579Z","iopub.status.idle":"2022-07-20T01:22:35.949407Z","shell.execute_reply.started":"2022-07-20T01:22:35.941542Z","shell.execute_reply":"2022-07-20T01:22:35.948356Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"X_train, X_val, Y_train, Y_val = train_test_split(X_train, Y_train, test_size = 0.1)","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:35.951074Z","iopub.execute_input":"2022-07-20T01:22:35.951917Z","iopub.status.idle":"2022-07-20T01:22:36.328727Z","shell.execute_reply.started":"2022-07-20T01:22:35.951883Z","shell.execute_reply":"2022-07-20T01:22:36.327741Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# CNN Model","metadata":{}},{"cell_type":"code","source":"model = Sequential()\n\nmodel.add(Conv2D(filters = 32, kernel_size = (5,5),padding = 'Same', \n                 activation ='relu', input_shape = (28,28,1)))\nmodel.add(BatchNormalization())\nmodel.add(Conv2D(filters = 32, kernel_size = (5,5),padding = 'Same', \n                 activation ='relu'))\nmodel.add(BatchNormalization())\nmodel.add(MaxPool2D(pool_size=(2,2)))\nmodel.add(Dropout(0.25))\n\nmodel.add(Conv2D(filters = 32, kernel_size = (3,3),padding = 'Same', \n                 activation ='relu', input_shape = (28,28,1)))\nmodel.add(BatchNormalization())\nmodel.add(Conv2D(filters = 32, kernel_size = (3,3),padding = 'Same', \n                 activation ='relu'))\nmodel.add(BatchNormalization())\nmodel.add(MaxPool2D(pool_size=(2,2)))\nmodel.add(Dropout(0.25))\n\nmodel.add(Flatten())\nmodel.add(Dense(256, activation = \"relu\"))\nmodel.add(Dropout(0.5))\nmodel.add(Dense(10, activation = \"softmax\"))","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:36.330121Z","iopub.execute_input":"2022-07-20T01:22:36.330483Z","iopub.status.idle":"2022-07-20T01:22:38.967932Z","shell.execute_reply.started":"2022-07-20T01:22:36.330448Z","shell.execute_reply":"2022-07-20T01:22:38.966153Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"optimizer = keras.optimizers.Nadam()","metadata":{"execution":{"iopub.status.busy":"2022-07-22T09:50:21.089899Z","iopub.execute_input":"2022-07-22T09:50:21.090275Z","iopub.status.idle":"2022-07-22T09:50:21.185271Z","shell.execute_reply.started":"2022-07-22T09:50:21.090197Z","shell.execute_reply":"2022-07-22T09:50:21.183245Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.compile(optimizer = optimizer, loss = \"categorical_crossentropy\", metrics=[\"accuracy\"])","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:38.977520Z","iopub.execute_input":"2022-07-20T01:22:38.978087Z","iopub.status.idle":"2022-07-20T01:22:38.995889Z","shell.execute_reply.started":"2022-07-20T01:22:38.978054Z","shell.execute_reply":"2022-07-20T01:22:38.994751Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Set a learning rate annealer\nlearning_rate_reduction = ReduceLROnPlateau(monitor='val_loss', \n                                            patience=3, \n                                            verbose=1, \n                                            factor=0.5, \n                                            min_lr=0.00001)","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:38.999986Z","iopub.execute_input":"2022-07-20T01:22:39.000259Z","iopub.status.idle":"2022-07-20T01:22:39.006138Z","shell.execute_reply.started":"2022-07-20T01:22:39.000234Z","shell.execute_reply":"2022-07-20T01:22:39.005036Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"epochs = 35 # Turn epochs to 30 to get 0.9967 accuracy\nbatch_size = 86","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:39.008943Z","iopub.execute_input":"2022-07-20T01:22:39.010576Z","iopub.status.idle":"2022-07-20T01:22:39.016258Z","shell.execute_reply.started":"2022-07-20T01:22:39.010549Z","shell.execute_reply":"2022-07-20T01:22:39.014979Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# With data augmentation to prevent overfitting (accuracy 0.99286)\n\ndatagen = ImageDataGenerator(\n        featurewise_center=False,  # set input mean to 0 over the dataset\n        samplewise_center=False,  # set each sample mean to 0\n        featurewise_std_normalization=False,  # divide inputs by std of the dataset\n        samplewise_std_normalization=False,  # divide each input by its std\n        zca_whitening=False,  # apply ZCA whitening\n        rotation_range=10,  # randomly rotate images in the range (degrees, 0 to 180)\n        zoom_range = 0.1, # Randomly zoom image \n        width_shift_range=0.1,  # randomly shift images horizontally (fraction of total width)\n        height_shift_range=0.1,  # randomly shift images vertically (fraction of total height)\n        horizontal_flip=False,  # randomly flip images\n        vertical_flip=False)  # randomly flip images\n\n\ndatagen.fit(X_train)","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:39.017521Z","iopub.execute_input":"2022-07-20T01:22:39.018353Z","iopub.status.idle":"2022-07-20T01:22:39.119186Z","shell.execute_reply.started":"2022-07-20T01:22:39.018321Z","shell.execute_reply":"2022-07-20T01:22:39.118174Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Fit the model\nhistory = model.fit(datagen.flow(X_train,Y_train, batch_size=batch_size),\n                              epochs = epochs, validation_data = (X_val,Y_val),\n                              verbose = 2, steps_per_epoch=X_train.shape[0] // batch_size,\n                              callbacks=[learning_rate_reduction])","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:22:39.120930Z","iopub.execute_input":"2022-07-20T01:22:39.121285Z","iopub.status.idle":"2022-07-20T01:23:18.216832Z","shell.execute_reply.started":"2022-07-20T01:22:39.121249Z","shell.execute_reply":"2022-07-20T01:23:18.215782Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Evaluation","metadata":{}},{"cell_type":"code","source":"# Plot the loss and accuracy curves for training and validation \nfig, ax = plt.subplots(2,1)\nax[0].plot(history.history['loss'], color='b', label=\"Training loss\")\nax[0].plot(history.history['val_loss'], color='r', label=\"validation loss\",axes =ax[0])\nlegend = ax[0].legend(loc='best', shadow=True)\n\nax[1].plot(history.history['accuracy'], color='b', label=\"Training accuracy\")\nax[1].plot(history.history['val_accuracy'], color='r',label=\"Validation accuracy\")\nlegend = ax[1].legend(loc='best', shadow=True)","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:23:18.218535Z","iopub.execute_input":"2022-07-20T01:23:18.218866Z","iopub.status.idle":"2022-07-20T01:23:18.586159Z","shell.execute_reply.started":"2022-07-20T01:23:18.218840Z","shell.execute_reply":"2022-07-20T01:23:18.585261Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Look at confusion matrix \n\ndef plot_confusion_matrix(cm, classes,\n                          normalize=False,\n                          title='Confusion matrix',\n                          cmap=plt.cm.Blues):\n    \"\"\"\n    This function prints and plots the confusion matrix.\n    Normalization can be applied by setting `normalize=True`.\n    \"\"\"\n    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n    plt.title(title)\n    plt.colorbar()\n    tick_marks = np.arange(len(classes))\n    plt.xticks(tick_marks, classes, rotation=45)\n    plt.yticks(tick_marks, classes)\n\n    if normalize:\n        cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n\n    thresh = cm.max() / 2.\n    for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n        plt.text(j, i, cm[i, j],\n                 horizontalalignment=\"center\",\n                 color=\"white\" if cm[i, j] > thresh else \"black\")\n\n    plt.tight_layout()\n    plt.ylabel('True label')\n    plt.xlabel('Predicted label')\n\n# Predict the values from the validation dataset\nY_pred = model.predict(X_val)\n# Convert predictions classes to one hot vectors \nY_pred_classes = np.argmax(Y_pred,axis = 1) \n# Convert validation observations to one hot vectors\nY_true = np.argmax(Y_val,axis = 1) \n# compute the confusion matrix\nconfusion_mtx = confusion_matrix(Y_true, Y_pred_classes) \n# plot the confusion matrix\nplot_confusion_matrix(confusion_mtx, classes = range(10)) ","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:23:18.587575Z","iopub.execute_input":"2022-07-20T01:23:18.588297Z","iopub.status.idle":"2022-07-20T01:23:19.448182Z","shell.execute_reply.started":"2022-07-20T01:23:18.588258Z","shell.execute_reply":"2022-07-20T01:23:19.447270Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Display some error results \n\n# Errors are difference between predicted labels and true labels\nerrors = (Y_pred_classes - Y_true != 0)\n\nY_pred_classes_errors = Y_pred_classes[errors]\nY_pred_errors = Y_pred[errors]\nY_true_errors = Y_true[errors]\nX_val_errors = X_val[errors]\n\ndef display_errors(errors_index,img_errors,pred_errors, obs_errors):\n    \"\"\" This function shows 6 images with their predicted and real labels\"\"\"\n    n = 0\n    nrows = 2\n    ncols = 3\n    fig, ax = plt.subplots(nrows,ncols,sharex=True,sharey=True)\n    for row in range(nrows):\n        for col in range(ncols):\n            error = errors_index[n]\n            ax[row,col].imshow((img_errors[error]).reshape((28,28)))\n            ax[row,col].set_title(\"Predicted label :{}\\nTrue label :{}\".format(pred_errors[error],obs_errors[error]))\n            n += 1\n\n# Probabilities of the wrong predicted numbers\nY_pred_errors_prob = np.max(Y_pred_errors,axis = 1)\n\n# Predicted probabilities of the true values in the error set\ntrue_prob_errors = np.diagonal(np.take(Y_pred_errors, Y_true_errors, axis=1))\n\n# Difference between the probability of the predicted label and the true label\ndelta_pred_true_errors = Y_pred_errors_prob - true_prob_errors\n\n# Sorted list of the delta prob errors\nsorted_dela_errors = np.argsort(delta_pred_true_errors)\n\n# Top 6 errors \nmost_important_errors = sorted_dela_errors[-6:]\n\n# Show the top 6 errors\ndisplay_errors(most_important_errors, X_val_errors, Y_pred_classes_errors, Y_true_errors)","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:23:19.449828Z","iopub.execute_input":"2022-07-20T01:23:19.450508Z","iopub.status.idle":"2022-07-20T01:23:19.989179Z","shell.execute_reply.started":"2022-07-20T01:23:19.450472Z","shell.execute_reply":"2022-07-20T01:23:19.988295Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# Prediction & Submission","metadata":{}},{"cell_type":"code","source":"# predict results\nresults = model.predict(test)\n\n# select the indix with the maximum probability\nresults = np.argmax(results,axis = 1)\n\nresults = pd.Series(results,name=\"Label\")","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:23:19.990414Z","iopub.execute_input":"2022-07-20T01:23:19.990851Z","iopub.status.idle":"2022-07-20T01:23:22.830748Z","shell.execute_reply.started":"2022-07-20T01:23:19.990813Z","shell.execute_reply":"2022-07-20T01:23:22.829760Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission = pd.concat([pd.Series(range(1,28001),name = \"ImageId\"),results],axis = 1)\n\nsubmission.to_csv(\"cnn_submission.csv\",index=False)","metadata":{"execution":{"iopub.status.busy":"2022-07-20T01:23:22.832129Z","iopub.execute_input":"2022-07-20T01:23:22.832592Z","iopub.status.idle":"2022-07-20T01:23:22.880009Z","shell.execute_reply.started":"2022-07-20T01:23:22.832555Z","shell.execute_reply":"2022-07-20T01:23:22.879134Z"},"trusted":true},"execution_count":null,"outputs":[]}]}