Question

我整天都在阅读帖子，试图找出如何从已保存的Tensorflow模型中进行预测，但是我无法在模型中使用任何示例。

我的模型调用了返回预测（输出）和损失的卷积神经网络：

`def my_training_task4(X_train, y_train, X_val, y_val...)

    N, height, width, channels =  X_train.shape
    with tf.name_scope('inputs'):
        xs = tf.placeholder(shape=[None, height, width, channels], dtype=tf.float32, name='xs')
        ys = tf.placeholder(shape=[None, ], dtype=tf.int64, name='ys')
        is_training = tf.placeholder(tf.bool, name='is_training')
        decay_tf = tf.placeholder(tf.float32, name='accum_lr_decay')

    #This is where I call the Convolution Network:
    output, loss = my_LeNet(xs, ys, is_training, etc...)

    iters = int(N / batch_size)


    #Here I call a Tensorflow Optimizer:
    step = train_step(loss, learning_rate*decay_tf, optimizer)

    #This just computes the error of the output:
    eve = evaluate(output, ys) 

    iter_total = 0
    epc = 0
    best_acc = 0
    cur_model_name = 'lenet_{}'.format(int(time.time()))

    with tf.Session() as sess:
        merge = tf.summary.merge_all()

        writer = tf.summary.FileWriter("log/{}".format(cur_model_name), sess.graph)
        saver = tf.train.Saver()
        extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
        sess.run([tf.global_variables_initializer()])

        decay = 1
        generator = ImageGenerator(X_train, y_train)
        for training_batch_x, training_batch_y in generator.next_batch_gen(batch_size, shuffle=True):
            if iter_total % iters == 0:
                print('epoch {}: new learning rate = {}'.format(epc+1, learning_rate*decay))
                epc += 1
                decay *= learning_decay

            output, cur_loss, update = sess.run([step, loss, extra_update_ops], feed_dict={xs: training_batch_x,
                                                                ys: training_batch_y,
                                                                is_training: True,
                                                                decay_tf: decay})
            if iter_total % 100 == 0:
                # do validation
                valid_eve, merge_result, update = sess.run([eve, merge, extra_update_ops], feed_dict={xs: X_val,
                                                                            ys: y_val,
                                                                            is_training: False})


                # when achieve the best validation accuracy, we store the model parameters (here we save the model):
                if valid_acc > best_acc:
                    best_acc = valid_acc
                    saver.save(sess, 'model/{}'.format(cur_model_name))


                    #save_path = saver.save(sess, 'model/{}.ckpt'.format(cur_model_name))
                    #print("Model saved in path: %s" % save_path)
            #decay *= learning_decay
            iter_total += 1
            if iter_total == iters * epoch:
                break

`

现在我想做的就是上传模型，并使用它对测试数据进行预测，但是我尝试了很多方法，但它们不起作用。我最近的人是这样的：

imported_meta = tf.train.import_meta_graph("model/lenet_1540941414.meta")

xs = tf.placeholder(shape=X_test.shape, dtype=tf.float32, name='xs')
is_training = tf.placeholder(tf.bool, name='is_training')

sess = tf.Session()
imported_meta.restore(sess, "model/lenet_1540941414")
P = sess.run([output], feed_dict={"xs:0": X_test, "is_training:0": False})

但是我得到一个错误，说xs不属于图形。

任何帮助将不胜感激！

从此Tensorflow模型预测

0 个答案: