ValueError: Cannot feed value of shape (100, 160) for Tensor 'Placeholder:0', which has shape '(?,_valueerror: cannot feed value of shape (100, 37) f

Python在使用Tensorflow过程中，运行程序出现如下错误：
ValueError: Cannot feed value of shape (100, 160) for Tensor ‘Placeholder:0’, which has shape '(?,16000)
错误的含义是TensorFlow无法为Tensor’占位符：0’提供形状值。

placeholder是TensorFlow的占位符节点，由placeholder方法创建，也是一种常量，是由用户在调用run方法是传递的，也可以将placeholder理解为一种形参。本例创建如下：

#占位符，通过为输入图像和目标输出类别创建节点，来开始构建计算图。
x = tf.placeholder("float", shape=[None, 16000]) #160*100
y_ = tf.placeholder("float", shape=[None, 4])
1
2
3

参数说明：

• dtype：数据类型，必填，默认为value的数据类型，传入参数为tensorflow下的枚举值（float32，float64…）
• shape：数据形状，选填，不填则随传入数据的形状自行变动，可以在多次调用中传入不同形状的数据
• name：常量名，选填，默认值不重复，根据创建顺序为（Placeholder，Placeholder_1，Placeholder_2…）

给模型喂数据，传参如下：

x_train,x_test,y_train,y_test = train_test_split(x_data,y_data,test_size=0.2,random_state=1)
......
    batch_x = x_train[i]
    if i%100 == 0:
        train_accuracy = accuracy.eval(feed_dict={x:batch_x, y_: batch_y, keep_prob: 1.0})
        print ("step {}, training accuracy {}".format(i, train_accuracy))
1
2
3
4
5
6

图片数据获取及数据源如下;

for j in [file_name for file_name in list][1]:
    for i in data_read(datapath + '/' + j):
        a = cv2.imread('./' + datapath + '/' + j + '/' + i + '.jpg', cv2.IMREAD_GRAYSCALE)
        sum.append([a, j])

y_data = [v[1] for v in sum]
x_data = [v[0] for v in sum]
x_data = np.array(x_data)
1
2
3
4
5
6
7
8

在调试程序过程中，出现了如下等错误提示：

ValueError: Cannot feed value of shape (10，100, 160) for Tensor ‘Placeholder:0’, which has shape ‘(?, 16000)’

ValueError: Cannot feed value of shape (100, 160) for Tensor ‘Placeholder:0’, which has shape ‘(?, 16000)’

ValueError: Cannot feed value of shape (16000,) for Tensor ‘Placeholder:0’, which has shape ‘(?, 16000)’

根本原因是给模型喂数据的形状（shape）不一致，解决方案如下：

    batch_x = x_train[i]
    batch_x = np.reshape(batch_x, (-1, 16000))
    if i%100 == 0:
        train_accuracy = accuracy.eval(feed_dict={x:batch_x, y_: batch_y, keep_prob: 1.0})
        print ("step {}, training accuracy {}".format(i, train_accuracy))    
1
2
3
4
5

使用Python numpy函数：reshape（），修改数组对象形状为（？，16000），batch_x = np.reshape(batch_x, (-1, 16000))。