Question

我使用张量流训练了自然网络，而在测试过程中发生了错误，这表明我训练的图和恢复的图不匹配。但是定义用于还原的网络结构和用于训练的网络的代码是相似的。

                x_data, y_data = incentiveAndResponse(childSequence[ridx, :], order)
                x_test, y_test = incentiveAndResponse(testChildSequence[ridx, :], order)
                # define placeholder for inputs to network
                xs = tf.placeholder(tf.float32, [None, order])
                ys = tf.placeholder(tf.float32, [None, 1])
                # add hidden layer 
                l1 = add_layer(xs, order, NoNode, activation_function=tf.nn.tanh)
                # l2 = add_layer(l1, NoNode, NoNode, activation_function=tf.nn.tanh)
                # l3 = add_layer(l2, NoNode, NoNode, activation_function=tf.nn.tanh)
                # add output layer
                prediction = add_layer(l1, NoNode, 1, activation_function=None)
                # the error between prediciton and real data
                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.initialize_all_variables()
                sess = tf.Session()
                sess.run(init)
                saver = tf.train.Saver()

                for i in range(1000):

                    sess.run(train_step, feed_dict={xs: x_data, ys: y_data})
                    if i % 50 == 0:
                        # to see the step improvement
                        print(sess.run(loss, feed_dict={xs: x_data, ys: y_data}))
                saver.save(sess, './my_net/save_net.ckpt')
                tf.reset_default_graph()

                # test
                # define placeholder for inputs to network
                xsnew = tf.placeholder(tf.float32, [None, order])
                ysnew = tf.placeholder(tf.float32, [None, 1])
                # add hidden layer 
                l1 = add_layer(xsnew, order, NoNode, activation_function=tf.nn.tanh)
                # l2 = add_layer(l1, NoNode, NoNode, activation_function=tf.nn.tanh)
                # l3 = add_layer(l2, NoNode, NoNode, activation_function=tf.nn.tanh)
                prediction = add_layer(l1, NoNode, 1, activation_function=None)
                saver = tf.train.Saver()
                with tf.Session() as sess:
                    sess.run(tf.global_variables_initializer())
                    saver.restore(sess, './my_net/save_net.ckpt')
                    res = sess.run(prediction, feed_dict={xsnew: x_test, ysnew: y_test})
                shutil.rmtree('./my_net')

这是ide的结果：

InvalidArgumentError（请参阅上面的回溯）：从中还原   检查点失败。这很可能是由于两者之间的不匹配   当前图和来自检查点的图。请确保   您尚未更改基于检查点的预期图形。   原始错误：

分配需要两个张量的形状匹配。 lhs shape = [2,3] rhs   shape = [2,2] [[节点保存/分配（在   F：/code/python/NewIdea/WaveletCnnInFCM.py：270）]]

训练后恢复模型的错误

0 个答案: