Tensorflow:ValueError:形状必须等于等级,但是为0和2

时间:2017-09-18 07:02:04

标签: tensorflow neural-network

我在乘法(x1,Wo1)时遇到形状误差。但我无法找到原因。 错误:ValueError:形状必须等于等级,但为0和2
从形状0与其他形状合并。 for' add_2 / x' (op:' Pack')输入形状:[],[20,1]。

    import tensorflow as tf
    import numpy as np
    import pandas as pd
    import math

    df1=pd.read_csv(r'C:\Ocean of knowledge\Acads\7th sem\UGP\datasets\xTrain.csv')
    df1 = df1.dropna()
    xTrain = df1.values


    df2 = pd.read_csv(r'C:\Ocean of knowledge\Acads\7th sem\UGP\datasets\yTrain.csv')
    df2 = df2.dropna()
    yTrain = df2.values

    sess=tf.Session()    
    saver = tf.train.import_meta_graph(r'C:\Ocean of knowledge\Acads\7th sem\UGP\NeuralNet\my_model.meta')
    saver.restore(sess,tf.train.latest_checkpoint('./'))


    graph = tf.get_default_graph()
    w1 = graph.get_tensor_by_name("input:0")
    feed_dict ={w1:xTrain1}
    op_to_restore = graph.get_tensor_by_name("hidden:0")
    h1 = sess.run(op_to_restore,feed_dict)
    print(h1)

    n_input1 = 20
    n_hidden1 = 1

    def weight_variable(shape):
        initial = tf.truncated_normal(shape, stddev=0.1)
        return tf.Variable(initial)

    def bias_variable(shape):
        initial = tf.constant(0.1, shape=shape)
        return tf.Variable(initial)

    x1 = tf.placeholder(tf.float32, shape=[])
    Wo1 = weight_variable([20,1])
    bo1 = bias_variable([1])
    y1 = tf.nn.tanh(tf.matmul((x1,Wo1)+ bo1),name="op_to_restore2_")

    y1_ = tf.placeholder("float", [None,n_hidden1], name="check_")
    meansq1 = tf.reduce_mean(tf.square(y1- y1_), name="hello_")
    train_step1 = tf.train.AdamOptimizer(0.005).minimize(meansq1)

    #init = tf.initialize_all_variables()

    init = tf.global_variables_initializer()
    sess.run(init)

    n_rounds1 = 100
    batch_size1 = 5
    n_samp1 = 350

    for i in range(n_rounds1+1):    
        sample1 = np.random.randint(n_samp1, size=batch_size1)
        batch_xs1 = h1[sample1][:]
        batch_ys1 = yTrain[sample1][:]
        sess.run(x1, feed_dict={x1: batch_xs1, y1_:batch_ys1})

1 个答案:

答案 0 :(得分:0)

tf.matmul((x1,Wo1)+ bo1a使用了tf.matmul(a,b),这是矩阵乘法运算。 此操作要求bx1都是矩阵(张量为> =张量的张量)。

在您的情况下,您将x1 = tf.placeholder(tf.float32, shape=[]) 与{/ 1>}定义相似

Wo1

Wo1 = weight_variable([20,1]) 定义为

x1

如您所见,[]不是矩阵,而是标量(形状为Project X is missing required library: 'somelib.jar'的张量)。

也许你正在寻找明智的乘法?这是tf.multiply的用途。