张量流-tf.where TypeError

Question

I wrote the following test code (a part of a much bigger code)

import tensorflow as tf

update_boolean = [True, False, True, False, True, True]


with tf.Session() as sess:
    op = tf.where(update_boolean, lambda: tf.train.AdamOptimizer(0.1), lambda: tf.no_op())

I get the following error

TypeError: Expected binary or unicode string, got <function <lambda> at 0x000000000118E400>

how can I fix this?

what I want to do is create a list\\tensor of operators (which are optimizer\\do nothing) given a certain condition (not using tf.cond because i want to apply this to a batch so I have a list\\tensor of booleans) * working in tensorflow 1.0

Answer 1

tf.train.AdamOptimizer(0.1) returns an Optimizer object, it is not clear how this relates to a no_op. I suggest starting by creating the optimizer and then conditioning on gradient updates instead, for instance by zeroing gradient contributions from the part of the batch wanted. tf.where does not work with operations, it provides indices for accessing tensors.