TensorFlow apply_gradients ValueError

Question

我在倒数第二行代码中收到以下错误：不确定如何继续。谁能给我一些见解。

ValueError: Tensor("Variable_20:0", shape=(8, 8, 4, 32), dtype=float32_ref) must be from the same graph as Tensor("Variable_20/RMSProp_1:0", shape=(8, 8, 4, 32), dtype=float32_ref).

代码如下：

optimizer = tf.train.RMSPropOptimizer(0.00025, 0.95, 0.95, 0.01) 
readout = tf.reduce_mean(tf.reduce_sum(tf.mul(l_readout,a), reduction_indices=1))
cost  = tf.reduce_mean(tf.square(tf.sub(y,readout)))
grads = optimizer.compute_gradients(cost, localW)
grad_vals = sess.run([g for (g,v) in grads], feed_dict = {a: val_a, y: val_y, s: val_s})

placeholder_gradients = []
for var in localW:
    placeholder_gradients.append( (tf.placeholder('float',shape=var.get_shape()) , var) ) 

feed_dict = {}
for i in range(len(placeholder_gradients)):
    feed_dict[placeholder_gradients[i]] = grad_vals[i]

apply_gradients = optimizer.apply_gradients(placeholder_gradients) #ERROR LINE
apply_gradients.run(feed_dict=feed_dict) 

Answer 1

这可能与使用示例中未显示的线程有关。在我进一步研究如何在同一张图上使用线程之前，我将撤回这个问题。