基于tensorflow搭建模块化神经网络示例

好的，以下是一个基于TensorFlow搭建模块化神经网络的示例代码：

import tensorflow as tf

# 定义输入层
def create_input_layer(input_size):
    return tf.placeholder(tf.float32, shape=[None, input_size], name='input')

# 定义隐层
def create_hidden_layer(input_layer, num_neurons, activation):
    input_size = int(input_layer.shape[1])
    stddev = 2.0 / tf.sqrt(input_size + num_neurons)
    weights = tf.Variable(tf.truncated_normal([input_size, num_neurons], stddev=stddev))
    biases = tf.Variable(tf.zeros([num_neurons]))
    layer = tf.matmul(input_layer, weights) + biases
    if activation:
        layer = activation(layer)
    return layer

# 定义输出层
def create_output_layer(input_layer, num_classes):
    input_size = int(input_layer.shape[1])
    stddev = 2.0 / tf.sqrt(input_size + num_classes)
    weights = tf.Variable(tf.truncated_normal([input_size, num_classes], stddev=stddev))
    biases = tf.Variable(tf.zeros([num_classes]))
    layer = tf.matmul(input_layer, weights) + biases
    return layer

# 定义模型
def build_model(input_size, hidden_layers, num_classes):
    input_layer = create_input_layer(input_size)
    hidden_layer = input_layer
    for num_neurons, activation in hidden_layers:
        hidden_layer = create_hidden_layer(hidden_layer, num_neurons, activation)
    output_layer = create_output_layer(hidden_layer, num_classes)
    return input_layer, output_layer

# 测试
input_size = 784
hidden_layers = [(256, tf.nn.relu), (128, tf.nn.relu)]
num_classes = 10
input_layer, output_layer = build_model(input_size, hidden_layers, num_classes)
print(input_layer)
print(output_layer)

这个示例代码定义了一个三层神经网络，包括一个输入层、两个隐层和一个输出层。其中，输入层和输出层都比较简单，而隐层则是根据输入层的大小、神经元个数和激活函数来创建的。同时，这个代码还支持模型的构建和测试。