{"cells":[{"metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load in \n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the \"../input/\" directory.\n# For example, running this (by clicking run or pressing Shift+Enter) will list the files in the input directory\n\nimport os\nimport subprocess\nimport matplotlib.pyplot as plt\nfrom keras.datasets import mnist\n%matplotlib inline\nprint(os.listdir(\"../input\"))\nprint(os.listdir(\"../../\"))\n# Any results you write to the current directory are saved as output.","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"0b0bbd56f25e1333350387f949407c27a4d0c576"},"cell_type":"code","source":"","execution_count":null,"outputs":[]},{"metadata":{"_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","trusted":true,"_kg_hide-input":true},"cell_type":"code","source":"#读取数据\ntrain_data = pd.read_csv('../input/train.csv')\n#添加额外数据，参考:https://www.kaggle.com/loveunk/kaggle-digit-recognizer-keras-cnn-100-accuracy/comments\n(x_train, y_train), (x_test, y_test) = mnist.load_data()\ntrain_data1 = np.concatenate([x_train, x_test], axis=0)\nlabels1 = np.concatenate([y_train, y_test], axis=0)\nlabels2 = train_data.label\nlabels = np.concatenate([labels1, labels2], axis=0)\ntrain_data2 = train_data.drop(columns='label')\nimages = np.concatenate([train_data1.reshape([-1,28*28]), train_data2.values], axis=0)\nprint(images.shape)\nprint(labels.shape)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"334fb6e2c3210c3b6ffe02bc0e61c76b3460ffc5"},"cell_type":"code","source":"#显示数据\nimages = images.reshape([-1,28,28,1])\nplt.figure(num='digit',figsize=(9,9))\nfor i in range(9):\n    plt.subplot(3,3,i+1) \n    plt.title(labels[i])\n    plt.imshow(np.squeeze(images[i,:,:,]))\nplt.show()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"e96393c1abf0bb72989346bd36004c9bb85836d1"},"cell_type":"code","source":"from keras.utils.np_utils import to_categorical\nfrom sklearn.model_selection import train_test_split\n#数据归一化\ntrain_data = images / 255.0\n#将labels编码成one-hot\nlabels = to_categorical(labels, num_classes = 10)\n# 拆分数据集为训练集和验证集\ntrain_data, val_data, train_label, val_label = train_test_split(train_data, labels, test_size = 0.1, random_state=2)\nprint(train_data.shape)\nprint(val_data.shape)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"70aa36343f8010a409a5e4b19b96c742449e3a90"},"cell_type":"code","source":"from keras.models import Sequential\nfrom keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D, BatchNormalization,\\\n                        Permute, TimeDistributed, Bidirectional,GRU\nfrom keras.optimizers import RMSprop\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom keras.callbacks import ReduceLROnPlateau\n#网络结构参数\nhidden_unit = 32\nkernel_size = 3\n\n#定义网络结构\nmodel = Sequential()\nmodel.add(Conv2D(filters = 64, kernel_size = (3,3),padding = 'Same', activation ='relu', input_shape = (28,28,1)))\nmodel.add(Conv2D(filters = 64, kernel_size = (3,3),padding = 'Same', activation ='relu'))\nmodel.add(BatchNormalization())\n\nmodel.add(MaxPool2D(pool_size=(2,2),strides=(2,2)))\nmodel.add(Dropout(0.25))\n\nmodel.add(Conv2D(filters = 128, kernel_size = (3,3),padding = 'Same', activation ='relu'))\nmodel.add(Conv2D(filters = 128, kernel_size = (3,3),padding = 'Same', activation ='relu'))\nmodel.add(BatchNormalization())\n\nmodel.add(MaxPool2D(pool_size=(2,2), strides=(2,2)))\nmodel.add(Dropout(0.25))\n\nmodel.add(Conv2D(filters = 256, kernel_size = (3,3),padding = 'Same', activation ='relu'))\nmodel.add(Conv2D(filters = 256, kernel_size = (3,3),padding = 'Same', activation ='relu'))\nmodel.add(BatchNormalization())\n\nmodel.add(MaxPool2D(pool_size=(1,2), strides=(1,2)))\nmodel.add(Dropout(0.25))\n\nmodel.add(TimeDistributed(Flatten()))\nmodel.add(Bidirectional(GRU(hidden_unit,return_sequences=True)))\nmodel.add(Flatten())\nmodel.add(Dense(256, activation = \"relu\"))\nmodel.add(BatchNormalization())\nmodel.add(Dropout(0.25))\nmodel.add(Dense(10, activation = \"softmax\"))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"56d8a79f80fc61a52f38faad94b3f2fc8391f7a5"},"cell_type":"code","source":"#显示网络结构图\nfrom keras.utils import plot_model\nplot_model(model, to_file='model.png', show_shapes=True, show_layer_names=True)\nfrom IPython.display import Image\nImage(\"model.png\")","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"57a4afa3145d5336842bde692b80078ba09d8157"},"cell_type":"code","source":"optimizer = RMSprop(lr=0.005, rho=0.9, epsilon=1e-08, decay=0.0)\nmodel.compile(optimizer = optimizer , loss = \"categorical_crossentropy\", metrics=[\"accuracy\"])\n# 设置学习速率\nlearning_rate = ReduceLROnPlateau(monitor='val_acc', \n                                            patience=3, \n                                            verbose=1, \n                                            factor=0.5, \n                                            min_lr=0.00001)\nepochs = 1\nbatch_size = 128\n#数据增强\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\ndatagen.fit(train_data)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"dcb326a55d60cdd085c4649df963f222f90685a3"},"cell_type":"code","source":"# 训练模型\nhistory = model.fit_generator(datagen.flow(train_data,train_label, batch_size=batch_size),\n                              epochs = epochs, validation_data = (val_data,val_label),\n                              verbose = 2, steps_per_epoch=train_data.shape[0] // batch_size\n                              , callbacks=[learning_rate])","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"7307118920aefa1e3d3f7e135fa100c1be84ab59"},"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['acc'], color='b', label=\"Training accuracy\")\nax[1].plot(history.history['val_acc'], color='r',label=\"Validation accuracy\")\nlegend = ax[1].legend(loc='best', shadow=True)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"3ecb3c3f02272fa68b315d74223d3643ec7e6bc6"},"cell_type":"code","source":"from sklearn.metrics import confusion_matrix\nimport itertools\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(val_data)\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(val_label,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)) ","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"91ecf91a2641cc4c3b5e789db3776b8d15a8f571"},"cell_type":"code","source":"#Display some error results \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 = val_data[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)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"2f101fb795fcf692d425d3c80ef5d482134d7d81"},"cell_type":"code","source":"#读取测试数据\ntest = pd.read_csv(\"../input/test.csv\")\ntest = test / 255.0\ntest = test.values.reshape(-1,28,28,1)\nresults = model.predict(test)\nresults = np.argmax(results,axis = 1)\nresults = pd.Series(results,name=\"Label\")","execution_count":null,"outputs":[]},{"metadata":{"trusted":true,"_uuid":"d4672e008ba07837d308cd08fd912f2ff167e960"},"cell_type":"code","source":"# 、\nsubmission = pd.concat([pd.Series(range(1,28001),name = \"ImageId\"),results],axis = 1)\nsubmission.to_csv(\"crnn_mnist_submission.csv\",index=False)\nprint(os.listdir('../input/'))","execution_count":null,"outputs":[]}],"metadata":{"kernelspec":{"display_name":"Python 3","language":"python","name":"python3"},"language_info":{"name":"python","version":"3.6.6","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"}},"nbformat":4,"nbformat_minor":1}