TensorFlow:使用图表前面的计算结果

时间:2016-06-09 21:17:19

标签: python tensorflow

这是一个小例子,我想根据计算结束时得到的值对批处理进行采样。

myArray

这是我定义获取批处理的函数的位置。

with tf.Session() as sess:

conditional_val = tf.Variable(0, trainable=False)

print("get labels")
images, labels = get_img_and_label(data_dir, batch_size, conditional_val)

print("infer")
logits = inference(images)

print("define loss")
loss = compute_loss(logits,labels)

for i range(1000):
    if i < 1:
        _, loss_value = sess.run([loss])
    else:
        _, loss_value = sess.run([loss],feed_dict={conditional_val: index})

    index = int(np.around(i*loss_value)[0])

它一直给我错误:

def get_img_and_labelS(data_dir,batch_size,conditional_val=0):
    with open(data_dir) as datafile:
        data = json.load(datafile)

    inputs = data.keys()
    input_names = []
    for i in range(len(inputs)):
        input_names.append(data_dir + inputs[i] + '.png')

    for f in input_names:
        if not tf.gfile.Exists(f):
            raise ValueError('Failed to find file: ' + f)

    images = []
    labels = []
    for i in range(len(input_names)):
    # Load images with PILLOW (PIL Library)
        my_img = Image.open(input_names[i])
        images.append(np.array(my_img))
        labels.append(data[inputs[i]][0])

    images = images[0*conditional_val:20*conditional_val]
    labels = labels[0*conditional_val:20*conditional_val]

    return(images,labels)

有什么好方法可以做我想做的事情吗? 谢谢!

0 个答案:

没有答案