Question

我正在尝试在Tensorflow中恢复我的模型。这就是我保存模型的方式：

ae = autoencoder(input_shape=[None, height, width, depth], conv_strides=[[1, stride1, stride1, 1], [1, stride2, stride2, 1]], n_filters=[1, num_filters, num_filters], filter_sizes=[size_filter, size_filter, size_filter], corruption=False, poolsize=2)

optimizer = tf.train.AdamOptimizer(learning_rate).minimize(ae['cost'])

# create a session to use the graph
init = tf.global_variables_initializer()
saver = tf.train.Saver()
with tf.Session() as sess:
    sess.run(init)
    # Network is trained here
    ...
    saver.save(sess, "model.ckpt")

然后我尝试使用此代码恢复它（在另一个文件中，在训练模型后，在单独的会话中）：

with tf.Session() as sess:
    saver = tf.train.import_meta_graph("model.ckpt.meta")
    saver.restore(sess, "model.ckpt")
    print("Model restored")
    ae = autoencoder(input_shape=[None, height, width, depth], conv_strides=[[1, stride1, stride1, 1], [1, stride2, stride2, 1]], n_filters=[1, num_filters, num_filters], filter_sizes=[size_filter, size_filter, size_filter], corruption=False, poolsize=2)
    # create stuff here to reconstruct images using the autoencoder
    ...
    recon = sess.run(ae['y'], feed_dict={ae['x']: batch})

打印出模型已恢复，但出现错误：
FailedPreconditionError：尝试使用未初始化的值
根据Tensorflow文档，您不必在恢复后初始化变量，所以我想它不会出错。有谁知道如何解决这一问题？我有一种感觉，我正在做一些非常愚蠢的事......

Answer 1

试试这个：

ae = autoencoder(input_shape=[None, height, width, depth], conv_strides=
[[1, stride1, stride1, 1], [1, stride2, stride2, 1]], n_filters=[1, num_filters, num_filters], filter_sizes=[size_filter, size_filter, size_filter], corruption=False, poolsize=2)    
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(ae['cost'])

saver = tf.train.Saver()

with tf.Session() as sess:
    saver.restore(sess, "model.ckpt")
    print("Model restored")
    # create stuff here to reconstruct images using the autoencoder
    ...
    recon = sess.run(ae['y'], feed_dict={ae['x']: batch})

Tensorflow恢复模型：尝试使用未初始化的值

1 个答案: