keras怎么改变维度_基于keras的手写数字识别及cifar10CNN图像分类实验讲解

(X_train, y_train), (X_test, y_test) &＃61; mnist.load_data()
print(&＃39;X_train shape:&＃39;, X_train.shape)
print(&＃39;X_test shape: &＃39;, X_test.shape)
print(&＃39;y_train shape:&＃39;,y_train.shape)
print(&＃39;y_test shape: &＃39;, y_test.shape)

# Import Numpy, keras and MNIST data
import numpy as np
import matplotlib.pyplot as pltfrom keras.datasets import mnist
from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation
from keras.utils import np_utils# Retrieve the training and test data
(X_train, y_train), (X_test, y_test) &＃61; mnist.load_data()
print(&＃39;X_train shape:&＃39;, X_train.shape)
print(&＃39;X_test shape: &＃39;, X_test.shape)
print(&＃39;y_train shape:&＃39;,y_train.shape)
print(&＃39;y_test shape: &＃39;, y_test.shape)# Visualizing the data
import matplotlib.pyplot as plt
%matplotlib inline# Function for displaying a training image by it&＃39;s index in the MNIST set
def display_digit(index):label &＃61; y_train[index].argmax(axis&＃61;0)image &＃61; X_train[index]plt.title(&＃39;Training data, index: %d, Label: %d&＃39; % (index, label))plt.imshow(image, cmap&＃61;&＃39;gray_r&＃39;)plt.show()# Display the first (index 0) training image
display_digit(0)X_train &＃61; X_train.reshape(60000, 784)
X_test &＃61; X_test.reshape(10000, 784)
X_train &＃61; X_train.astype(&＃39;float32&＃39;)
X_test &＃61; X_test.astype(&＃39;float32&＃39;)
X_train /&＃61; 255
X_test /&＃61; 255
print("Training matrix shape", X_train.shape)
print("Testing matrix shape", X_test.shape)#One Hot encoding of labels.
from keras.utils.np_utils import to_categorical
print(y_train.shape)
y_train &＃61; to_categorical(y_train, 10)
y_test &＃61; to_categorical(y_test, 10)
print(y_train.shape)# Define the neural network
def build_model():model &＃61; Sequential()model.add(Dense(512, input_shape&＃61;(784,)))model.add(Activation(&＃39;relu&＃39;)) # An "activation" is just a non-linear function applied to the output# of the layer above. Here, with a "rectified linear unit",# we clamp all values below 0 to 0.model.add(Dropout(0.2)) # Dropout helps protect the model from memorizing or "overfitting" the training datamodel.add(Dense(512))model.add(Activation(&＃39;relu&＃39;))model.add(Dropout(0.2))model.add(Dense(10))model.add(Activation(&＃39;softmax&＃39;)) # This special "softmax" activation among other things,# ensures the output is a valid probaility distribution, that is# that its values are all non-negative and sum to 1.return model# Build the model
model &＃61; build_model()model.compile(optimizer&＃61;&＃39;rmsprop&＃39;,loss&＃61;&＃39;categorical_crossentropy&＃39;,metrics&＃61;[&＃39;accuracy&＃39;])# Training
model.fit(X_train, y_train, batch_size&＃61;128, nb_epoch&＃61;4, verbose&＃61;1,validation_data&＃61;(X_test, y_test))# Compare the labels that our model predicts with the actual labelsscore &＃61; model.evaluate(X_test, y_test, batch_size&＃61;32, verbose&＃61;1,sample_weight&＃61;None)
# Print out the result
print(&＃39;Test score:&＃39;, score[0])
print(&＃39;Test accuracy:&＃39;, score[1])

import keras
from keras.datasets import cifar10# load the pre-shuffled train and test data
# link to dataset: https://www.cs.toronto.edu/~kriz/cifar.html
(x_train, y_train), (x_test, y_test) &＃61; cifar10.load_data()
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inlinefig &＃61; plt.figure(figsize&＃61;(20,5))
for i in range(36):ax &＃61; fig.add_subplot(3, 12, i &＃43; 1, xticks&＃61;[], yticks&＃61;[])ax.imshow(np.squeeze(x_train[i]))

# rescale [0,255] --> [0,1]
x_train &＃61; x_train.astype(&＃39;float32&＃39;)/255
x_test &＃61; x_test.astype(&＃39;float32&＃39;)/255
x_train[0]

y_train[0]

from keras.utils import np_utils# break training set into training and validation sets
(x_train, x_valid) &＃61; x_train[5000:], x_train[:5000]
(y_train, y_valid) &＃61; y_train[5000:], y_train[:5000]# one-hot encode the labels
num_classes &＃61; len(np.unique(y_train))
y_train &＃61; keras.utils.to_categorical(y_train, num_classes)
y_test &＃61; keras.utils.to_categorical(y_test, num_classes)
y_valid &＃61; keras.utils.to_categorical(y_valid, num_classes)# print shape of training set
print(&＃39;x_train shape:&＃39;, x_train.shape)# print number of training, validation, and test images
print(x_train.shape[0], &＃39;train samples&＃39;)
print(x_test.shape[0], &＃39;test samples&＃39;)
print(x_valid.shape[0], &＃39;validation samples&＃39;)

from keras.preprocessing.image import ImageDataGenerator# create and configure augmented image generator
datagen_train &＃61; ImageDataGenerator(width_shift_range&＃61;0.1, # randomly shift images horizontally (10% of total width)height_shift_range&＃61;0.1, # randomly shift images vertically (10% of total height)horizontal_flip&＃61;True) # randomly flip images horizontally# fit augmented image generator on data
datagen_train.fit(x_train)#在这里我们可以看到增强的图像。我们从训练数据中获取10张图像&＃xff0c;并查看原始数据和增强数据。尝试将原始图像与增强后的图像进行匹配。
#哪些是水平翻转的?
#哪些是水平和垂直shiffted的?import matplotlib.pyplot as pltnum_to_view &＃61; 10# take subset of training data
x_train_subset &＃61; x_train[:num_to_view]# visualize subset of training data
fig &＃61; plt.figure(figsize&＃61;(20,2))
for i in range(0, len(x_train_subset)):ax &＃61; fig.add_subplot(1, num_to_view, i&＃43;1)ax.imshow(x_train_subset[i])
fig.suptitle(&＃39;Subset of Original Training Images&＃39;, fontsize&＃61;20)
plt.show()# visualize augmented images
fig &＃61; plt.figure(figsize&＃61;(20,2))
for x_batch in datagen_train.flow(x_train_subset, batch_size&＃61;12):for i in range(0, num_to_view):ax &＃61; fig.add_subplot(1, num_to_view, i&＃43;1)ax.imshow(x_batch[i])fig.suptitle(&＃39;Augmented Images&＃39;, fontsize&＃61;20)plt.show()break;

from keras import backend as K
K.clear_session()from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropoutmodel &＃61; Sequential()
model.add(Conv2D(filters&＃61;16, kernel_size&＃61;2, padding&＃61;&＃39;same&＃39;, activation&＃61;&＃39;relu&＃39;, input_shape&＃61;(32, 32, 3)))
model.add(MaxPooling2D(pool_size&＃61;2))
model.add(Conv2D(filters&＃61;32, kernel_size&＃61;2, padding&＃61;&＃39;same&＃39;, activation&＃61;&＃39;relu&＃39;))
model.add(MaxPooling2D(pool_size&＃61;2))
model.add(Conv2D(filters&＃61;64, kernel_size&＃61;2, padding&＃61;&＃39;same&＃39;, activation&＃61;&＃39;relu&＃39;))
model.add(MaxPooling2D(pool_size&＃61;2))
model.add(Dropout(0.3))
model.add(Flatten())
model.add(Dense(500, activation&＃61;&＃39;relu&＃39;))
model.add(Dropout(0.4))
model.add(Dense(10, activation&＃61;&＃39;softmax&＃39;))model.summary()

# compile the model
model.compile(loss&＃61;&＃39;categorical_crossentropy&＃39;, optimizer&＃61;&＃39;rmsprop&＃39;, metrics&＃61;[&＃39;accuracy&＃39;])

from keras.callbacks import ModelCheckpoint batch_size &＃61; 32
epochs &＃61; 10# train the model
checkpointer &＃61; ModelCheckpoint(filepath&＃61;&＃39;aug_model.weights.best.hdf5&＃39;, verbose&＃61;1, save_best_only&＃61;True)# datagren.flow makes batches of augmented images. You provide it with the
# training images and labels along with the number of images you want to be
# generated in a batch.
model.fit_generator(datagen_train.flow(x_train, y_train, batch_size&＃61;batch_size),steps_per_epoch&＃61;x_train.shape[0] // batch_size,epochs&＃61;epochs, verbose&＃61;2, callbacks&＃61;[checkpointer],validation_data&＃61;(x_valid, y_valid),validation_steps&＃61;x_valid.shape[0] // batch_size)

# load the weights that yielded the best validation accuracy
model.load_weights(&＃39;aug_model.weights.best.hdf5&＃39;)

# evaluate and print test accuracy
score &＃61; model.evaluate(x_test, y_test, verbose&＃61;0)
print(&＃39;n&＃39;, &＃39;Test accuracy:&＃39;, score[1])

# get predictions on the test set
y_hat &＃61; model.predict(x_test)# define text labels (source: https://www.cs.toronto.edu/~kriz/cifar.html)
cifar10_labels &＃61; [&＃39;airplane&＃39;, &＃39;automobile&＃39;, &＃39;bird&＃39;, &＃39;cat&＃39;, &＃39;deer&＃39;, &＃39;dog&＃39;, &＃39;frog&＃39;, &＃39;horse&＃39;, &＃39;ship&＃39;, &＃39;truck&＃39;]

# plot a random sample of test images, their predicted labels, and ground truth
fig &＃61; plt.figure(figsize&＃61;(20, 8))
for i, idx in enumerate(np.random.choice(x_test.shape[0], size&＃61;32, replace&＃61;False)):ax &＃61; fig.add_subplot(4, 8, i &＃43; 1, xticks&＃61;[], yticks&＃61;[])ax.imshow(np.squeeze(x_test[idx]))pred_idx &＃61; np.argmax(y_hat[idx])true_idx &＃61; np.argmax(y_test[idx])ax.set_title("{} ({})".format(cifar10_labels[pred_idx], cifar10_labels[true_idx]),color&＃61;("green" if pred_idx &＃61;&＃61; true_idx else "red"))