# coding: utf-8# # TensorFlow卷积神经网络(CNN)示例 - 高级API# ### Convolutional Neural Network Example - tf.layers API# ## CNN网络结构图示# # ![CNN](http://personal.ie.cuhk.edu.hk/~ccloy/project_target_code/images/fig3.png)# # ## MNIST数据库# # ![MNIST Dataset](http://neuralnetworksanddeeplearning.com/images/mnist_100_digits.png)# # More info: http://yann.lecun.com/exdb/mnist/from__future__importdivision, print_function, absolute_import# Import MNIST data，MNIST数据集导入fromtensorflow.examples.tutorials.mnistimportinput_datamnist=input_data.read_data_sets("/tmp/data/", one_hot=False)
importtensorflowastfimportmatplotlib.pyplotaspltimportnumpyasnp# Training Parameters，超参数learning_rate=0.001# 学习率num_steps=2000# 训练步数batch_size=128# 训练数据批的大小# Network Parameters，网络参数num_input=784# MNIST数据输入 (img shape: 28*28)num_classes=10# MNIST所有类别 (0-9 digits)dropout=0.75# Dropout, probability to keep units，保留神经元相应的概率# Create the neural network，创建深度神经网络defconv_net(x_dict, n_classes, dropout, reuse, is_training):
# Define a scope for reusing the variables，确定命名空间withtf.variable_scope('ConvNet', reuse=reuse):
# TF Estimator类型的输入为像素x=x_dict['images']
# MNIST数据输入格式为一位向量，包含784个特征 (28*28像素)# 用reshape函数改变形状以匹配图像的尺寸 [高 x 宽 x 通道数]# 输入张量的尺度为四维: [(每一)批数据的数目, 高，宽，通道数]x=tf.reshape(x, shape=[-1, 28, 28, 1])
# 卷积层，32个卷积核，尺寸为5x5conv1=tf.layers.conv2d(x, 32, 5, activation=tf.nn.relu)
# 最大池化层，步长为2，无需学习任何参量conv1=tf.layers.max_pooling2d(conv1, 2, 2)
# 卷积层，32个卷积核，尺寸为5x5conv2=tf.layers.conv2d(conv1, 64, 3, activation=tf.nn.relu)
# 最大池化层，步长为2，无需学习任何参量conv2=tf.layers.max_pooling2d(conv2, 2, 2)
# 展开特征为一维向量，以输入全连接层fc1=tf.contrib.layers.flatten(conv2)
# 全连接层fc1=tf.layers.dense(fc1, 1024)
# 应用Dropout (训练时打开，测试时关闭)fc1=tf.layers.dropout(fc1, rate=dropout, training=is_training)
# 输出层，预测类别out=tf.layers.dense(fc1, n_classes)
returnout# 确定模型功能 (参照TF Estimator模版)defmodel_fn(features, labels, mode):
# 构建神经网络# 因为dropout在训练与测试时的特性不一，我们此处为训练和测试过程创建两个独立但共享权值的计算图logits_train=conv_net(features, num_classes, dropout, reuse=False, is_training=True)
logits_test=conv_net(features, num_classes, dropout, reuse=True, is_training=False)
# 预测pred_classes=tf.argmax(logits_test, axis=1)
pred_probas=tf.nn.softmax(logits_test)
ifmode==tf.estimator.ModeKeys.PREDICT:
returntf.estimator.EstimatorSpec(mode, predictions=pred_classes)
# 确定误差函数与优化器loss_op=tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
logits=logits_train, labels=tf.cast(labels, dtype=tf.int32)))
optimizer=tf.train.AdamOptimizer(learning_rate=learning_rate)
train_op=optimizer.minimize(loss_op, global_step=tf.train.get_global_step())
# 评估模型精确度acc_op=tf.metrics.accuracy(labels=labels, predictions=pred_classes)
# TF Estimators需要返回EstimatorSpecestim_specs=tf.estimator.EstimatorSpec(
mode=mode,
predictions=pred_classes,
loss=loss_op,
train_op=train_op,
eval_metric_ops={'accuracy': acc_op})
returnestim_specsif__name__=="__main__":
# 构建Estimatormodel=tf.estimator.Estimator(model_fn)
# 确定训练输入函数input_fn=tf.estimator.inputs.numpy_input_fn(
x={'images': mnist.train.images}, y=mnist.train.labels,
batch_size=batch_size, num_epochs=None, shuffle=True)
# 开始训练模型model.train(input_fn, steps=num_steps)
# 评判模型# 确定评判用输入函数input_fn=tf.estimator.inputs.numpy_input_fn(
x={'images': mnist.test.images}, y=mnist.test.labels,
batch_size=batch_size, shuffle=False)
model.evaluate(input_fn)
# 预测单个图像n_images=4# 从数据集得到测试图像test_images=mnist.test.images[:n_images]
# 准备输入数据input_fn=tf.estimator.inputs.numpy_input_fn(
x={'images': test_images}, shuffle=False)
# 用训练好的模型预测图片类别preds=list(model.predict(input_fn))
# 可视化显示foriinrange(n_images):
plt.imshow(np.reshape(test_images[i], [28, 28]), cmap='gray')
plt.show()
print("Model prediction:", preds[i])