TensorFlow 遇到的MatMul bug

python版本             3.6.3

TensorFlow版本     1.2.1

用写TensorFlow的时候出现了一个bug：

TypeError: Input 'b' of 'MatMul' Op has type float32 that does not match type float64 of argument 'a'.

数据类型错误：MatMul里输入的数据类型为‘float32’的‘b’无法和数据类型为‘float64’的‘a’匹配。

1.看一下添加层内部matmul里的情况：

Wx_plus_b = tf.matmul(inputs,Weights) + biases

2.那么inputs应该就是底层实现里的‘a’，Weights为‘b’，

沿着参数传递找到数据源头：

x_data = np.linspace(-1,1,300)[:,np.newaxis]

3.在代码后方加入一个类型转换：

x_data = np.linspace(-1,1,300)[:,np.newaxis].astype('float32')

F5

Well done.

源码：

import tensorflow as tf

import numpy as np

#定义添加层

def add_layer(inputs, in_size, out_size,

    activation_function = None):

    Weights = tf.Variable(tf.random_normal([in_size, out_size]))

    biases = tf.Variable(tf.zeros([1,out_size])+0.1)

    Wx_plus_b = tf.matmul(inputs,Weights) + biases

    if activation_function is None:

    outputs = Wx_plus_b

    else:

    outputs = activation_function(Wx_plus_b)

    return outputs

#生成数据

x_data = np.linspace(-1,1,300)[:,np.newaxis]

noise = np.random.normal(0,0.05,x_data.shape)

y_data = np.square(x_data) - 0.5 + noise

#建立空间

xs = tf.placeholder(tf.float32,[None,1])

ys = tf.placeholder(tf.float32,[None,1])

#添加层

l1 = add_layer(x_data,1,10,activation_function = tf.nn.relu)

prediction = add_layer(l1,10,1,activation_function = None)

#设置参数

loss = tf.reduce_mean(tf.reduce_sum(tf.square(ys-prediction),

reduction_indices = [1]))

train_step = tf.train.GradientDescentOptimizer(0.1).minimize(loss)

init = tf.global_variables_initializer()

#会话控制

sess = tf.Session()

sess.run(init)

for i in range(1000):

    sess.run(train_step,feed_dict = {xs:x_data,ys:y_data})

    if i %50 ==0:

        print(sess.run(loss,feed_dict={xs:x_data,ys:y_data}))