不允许迭代 `tf.Tensor`

编程问答 2022-08-26

问题描述

我正在尝试将此函数用于@tf.function 装饰：

h 和 h2 是形状为 [3,3] 的张量

def fn(h,i):
    print(h[i])
    return h[i]

tensor = [fn(h,i) for i in tf.range(tf.cast(tf.shape(h)[0],tf.int32)) if  tf.reduce_all(tf.equal(h[i],h2[i])) ]
tf.print(tensor)

但是我遇到了这个错误：

main_coat_rds.py:139 train_step  *
        pseudo_label_1,images_discard_rede1=predict_aug_images(rede_2,rede_1,img_rede1_aug_1,img_rede1_aug_2,img_rede1_aug_3,img_rede1_aug_4,img_rede1_aug_5,img_rede1_aug_6,img_rede1_aug_7,img_rede1_aug_8,images_discard_rede1,Correct_labels)
    /vitor/codigo_noise_label/codigo_rds/utils_loss_function.py:289 predict_aug_images  *
        pred_match = [get_value_labels(all_predics,i) for i in tf.range(tf.cast(tf.shape(all_predics)[0],tf.int32)) if  tf.reduce_all(tf.equal(all_predics[i],all_predics_aj[i])) ]
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py:503 __iter__
        self._disallow_iteration()
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py:496 _disallow_iteration
        self._disallow_when_autograph_enabled("iterating over `tf.Tensor`")
    /usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py:474 _disallow_when_autograph_enabled
        " indicate you are trying to use an unsupported feature.".format(task))

    OperatorNotAllowedInGraphError: iterating over `tf.Tensor` is not allowed: AutoGraph did convert this function. This might indicate you are trying to use an unsupported feature.

还有什么方法可以做到？

解决方法

每当您想根据条件选择张量的某个部分时，一个不错的选择是使用 tf.gather 和 tf.where 的组合。

例如，要选择 h 和 h2 之间相等的行，可以使用：

tf.gather_nd(h,tf.where(tf.reduce_all(tf.equal(h,h2),axis=1)))
pythontensortensortensorflowtensorflowtensorflowtensorflow2