Question

我使用tf.data.dataset将数据导入模型。我创建了一个简单的可重现代码来展示这个想法。我保存训练好的模型（请参考下面的代码），一旦我恢复模型在测试数据上运行它，我得到错误，迭代器尚未初始化。有关详细信息，请参阅以下错误：

FailedPreconditionError (see above for traceback): GetNext() failed 
because the iterator has not been initialized. Ensure that you have 
run the initializer operation for this iterator before getting the 
next element.
[[Node: IteratorGetNext = IteratorGetNext[output_shapes=[[?,10], 
[?,1]], output_types=[DT_FLOAT, DT_FLOAT], 
_device="/job:localhost/replica:0/task:0/device:CPU:0"](Iterator)]]
 [[Node: IteratorGetNext/_39 = _Recv[client_terminated=false, 
 recv_device="/job:localhost/replica:0/task:0/device:GPU:0", 
 send_device="/job:localhost/replica:0/task:0/device:CPU:0", 
 send_device_incarnation=1, tensor_name="edge_7_IteratorGetNext", 
 tensor_type=DT_FLOAT, 
 _device="/job:localhost/replica:0/task:0/device:GPU:0"]()]]

我该如何解决这个问题？这是可重现的代码：

import tensorflow as tf
import os
import numpy as np
import math

features=np.random.randn(100,10)
features_test=np.random.randn(10,10)
y=np.random.randn(100,1)
y_test=np.random.randn(10,1)
feature_size=features.shape[1]
state_size=5
learning_rate=0.001


graph = tf.Graph()
with graph.as_default():

batch_size_tensor = tf.placeholder(tf.int64,name="Batch_tensor")

X,Y = tf.placeholder(tf.float32,   
[None,feature_size],"X"),tf.placeholder(tf.float32,[None,1],name="Y")

dataset =tf.data.Dataset.from_tensor_slices((X,Y)).batch(batch_size_tensor).repeat()
iter = dataset.make_initializable_iterator()
x_inputs,y_outputs = iter.get_next()

Wx = tf.Variable(tf.truncated_normal([feature_size, state_size], stddev=2.0 / math.sqrt(state_size)),name="Visual_weights_layer1")
bx= tf.Variable(tf.zeros([state_size]),name="Visual_bias_layer1")
x_hidden_state=tf.matmul(x_inputs, Wx)+bx
x_hidden_state = tf.contrib.layers.batch_norm(x_hidden_state, epsilon=1e-5)
vx=tf.nn.relu(x_hidden_state)

W_final = tf.Variable(tf.truncated_normal([state_size, 1], stddev=2.0 / math.sqrt(state_size)),name="FinalLayer_weights")
by=tf.Variable(tf.zeros([1]),name="FinalLayer_bias")
predictions = tf.add(tf.matmul(vx, W_final), by,name="preds")

loss = tf.losses.mean_squared_error(y_outputs,predictions)
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(loss)
init = tf.global_variables_initializer()
saver = tf.train.Saver()


num_steps=100
batch_size=1

saver_path_model='tmp/testmodel'
export_path_model='tmp/testmodel.meta'



with tf.Session(graph=graph) as sess:
 sess.run(init)

 sess.run(iter.initializer, feed_dict={X: features, Y: y, 
 batch_size_tensor: batch_size})
 print('initialized.')


 for step in range(num_steps):

  _, loss_val = sess.run([optimizer, loss])

  print (loss_val)

 saver.save(sess, saver_path_model)
 saver.export_meta_graph(filename=export_path_model)


sess = tf.Session()
new_saver = tf.train.import_meta_graph(export_path_model)
new_saver.restore(sess, saver_path_model)
graph = tf.get_default_graph()

feed = {"X:0": features_test,"Y:0": y_test}
predictions_test = sess.run(["preds:0"], feed_dict=feed)

Answer 1

我保存了我的模型如下

if (getTokenWithPostMethod() == true)
    {
        lblResult.text = "yes, we get token: " + g_token
    }
    else
    {
        lblResult.text = "there is an error, please try again later"
    }

然后加载它

saver = tf.train.Saver()

with tf.Session() as session:
    session.run(tf.global_variables_initializer())
...
    # after all training
    save_path = saver.save(session, "logs/trained_model.ckpt")
    print("Model saved: {}".format(save_path))

官方文档有更多的例子 https://www.tensorflow.org/api_docs/python/tf/train/Saver

如何使用tf.data.dataset保存和恢复模型以导入数据？

1 个答案: