MomentumOptimizer-Fehler: Versuch, nicht initialisierten Wert zu verwenden Variable_2/Momentum

Question

Ich lerne TensorFlow. Ich habe versucht, tf.train.MomentumOptimizer aber ich bekam die folgende Fehlermeldung:

Traceback (most recent call last): 
    File "relu.py", line 98, in <module> 
    learner.run(stop=0.01, print_epoch=True) 
    File "relu.py", line 70, in run 
    self.sess.run(train_step, feed_dict={self.x: batch_xs, self.y_: batch_ys}) 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 767, in run 
    run_metadata_ptr) 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 965, in _run 
    feed_dict_string, options, run_metadata) 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 1015, in _do_run 
    target_list, options, run_metadata) 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/client/session.py", line 1035, in _do_call 
    raise type(e)(node_def, op, message) 
tensorflow.python.framework.errors_impl.FailedPreconditionError: Attempting to use uninitialized value Variable_2/Momentum 
    [[Node: Momentum/update_Variable_2/ApplyMomentum = ApplyMomentum[T=DT_FLOAT, _class=["loc:@Variable_2"], use_locking=false, use_nesterov=false, _device="/job:localhost/replica:0/task:0/cpu:0"](Variable_2, Variable_2/Momentum, Momentum/learning_rate, gradients/add_1_grad/tuple/control_dependency_1, Momentum/momentum)]] 

Caused by op u'Momentum/update_Variable_2/ApplyMomentum', defined at: 
    File "relu.py", line 98, in <module> 
    learner.run(stop=0.01, print_epoch=True) 
    File "relu.py", line 55, in run 
    train_step = self.optimizer.minimize(self.cross_entropy) 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/training/optimizer.py", line 289, in minimize 
    name=name) 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/training/optimizer.py", line 413, in apply_gradients 
    update_ops.append(processor.update_op(self, grad)) 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/training/optimizer.py", line 61, in update_op 
    return optimizer._apply_dense(g, self._v) # pylint: disable=protected-access 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/training/momentum.py", line 69, in _apply_dense 
    use_nesterov=self._use_nesterov).op 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/training/gen_training_ops.py", line 348, in apply_momentum 
    use_nesterov=use_nesterov, name=name) 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/op_def_library.py", line 763, in apply_op 
    op_def=op_def) 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 2327, in create_op 
    original_op=self._default_original_op, op_def=op_def) 
    File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/ops.py", line 1226, in __init__ 
    self._traceback = _extract_stack() 

FailedPreconditionError (see above for traceback): Attempting to use uninitialized value Variable_2/Momentum 
    [[Node: Momentum/update_Variable_2/ApplyMomentum = ApplyMomentum[T=DT_FLOAT, _class=["loc:@Variable_2"], use_locking=false, use_nesterov=false, _device="/job:localhost/replica:0/task:0/cpu:0"](Variable_2, Variable_2/Momentum, Momentum/learning_rate, gradients/add_1_grad/tuple/control_dependency_1, Momentum/momentum)]] 

Und folgende ist mein Code:

import time 
import numpy as np 
import tensorflow as tf 
import tensorflow.examples.tutorials.mnist.input_data as input_data 

class ReluMnistNet: 

    def __init__(self, optimizer=None): 
     self.varlist = [] 
     self.optimizer = optimizer or tf.train.GradientDescentOptimizer(0.01) 

     # fetch dataset 

     self.mnist = input_data.read_data_sets("MNIST_data/", one_hot=True) 

     # prepare environment 
     layers = [ 100 ] 
     input_layer = 784 
     output_layer = 10 

     self.x = tf.placeholder(tf.float32, [None, input_layer]) 
     last_layer = input_layer 
     y = self.x 
     for layer in layers: 
      b = tf.Variable(tf.zeros([layer])) 
      self.varlist.append(b) 
      W = tf.Variable(tf.random_normal([last_layer,layer], stddev=0.01)) 
      self.varlist.append(W) 
      y = tf.nn.relu(tf.matmul(y,W)) + b 
      last_layer = layer 
     b = tf.Variable(tf.zeros([output_layer])) 
     self.varlist.append(b) 
     W = tf.Variable(tf.random_normal([last_layer,output_layer], stddev=0.01)) 
     self.varlist.append(W) 
     self.y = tf.matmul(y,W) + b 

     self.y_ = tf.placeholder(tf.float32, [None, 10]) 

     self.cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=self.y, labels=self.y_)) 

    def prepare(self): 
     # init = tf.initialize_variables(self.varlist) 
     init = tf.initialize_all_variables() 

     self.sess = tf.Session() 
     self.sess.run(init) 

    def run(self, batch_size=100, stop=0.001, print_epoch=False): 
     mnist = self.mnist 
     data_size = mnist.train.images.shape[0] 

     last_accuracy = 0 

     accuracy_history = [] 
     train_step = self.optimizer.minimize(self.cross_entropy) 

     time1 = time.time() 
     for i in range(10000): 
      for j in range(data_size/batch_size): 
       # random batch 
       batch_idx = np.arange(data_size) 
       np.random.shuffle(batch_idx) 
       batch_idx = batch_idx[0:batch_size] 
       batch_xs = mnist.train.images[batch_idx] 
       batch_ys = mnist.train.labels[batch_idx] 
       # ordered batch 
       # start = j * batch_size 
       # end = (j+1) * batch_size 
       # batch_xs, batch_ys = mnist.train.images[start:end], mnist.train.labels[start:end] 
       self.sess.run(train_step, feed_dict={self.x: batch_xs, self.y_: batch_ys}) 

      # test the accuracy 
      correct_prediction = tf.equal(tf.argmax(self.y,1), tf.argmax(self.y_,1)) 
      accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) 
      accuracy = self.sess.run(accuracy, feed_dict = {self.x: mnist.test.images, self.y_: mnist.test.labels}) 
      accuracy_history.append(accuracy) 
      if print_epoch: 
       print i, accuracy 
      if last_accuracy != 0 and abs(last_accuracy-accuracy) < stop: 
       break 
      last_accuracy = accuracy 

     time2 = time.time() 

     return accuracy_history, (time2-time1) 

    def close(self): 
     if not (self.sess is None): 
      self.sess.close() 
      self.sess = None 

if __name__ == '__main__': 
    learner = ReluMnistNet() 
    # learner.optimizer = tf.train.GradientDescentOptimizer(0.01) 
    learner.optimizer = tf.train.MomentumOptimizer(0.01, momentum=0.9) 
    for i in range(10): 
     learner.prepare() 
     learner.run(stop=0.01, print_epoch=True) 
     learner.close() 

Es scheint wie eine Variable mit dem Namen Momentum ist nicht initialisiert? Durch Aufruf von learner.prepare() habe ich jedoch tf.initialize_all_variables() aufgerufen. Mehr noch, ich habe keine Variable mit dem Namen Momentum. Warum passiert das?

Answer 1

In Ihrem Code Sie anrufen minimize nach globalen Variablen

statt Initialisieren Sie tun müssen:

self.cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=self.y, labels=self.y_)) 
self.optimize = self.optimizer.minimize(self.cross_entropy) 

und in Lauffunktion statt

train_step = self.optimizer.minimize(self.cross_entropy) 

sollten Sie rufen

train_step = self.optimize 

P.S Momentun ist der Standardname für den MomentumOptimizer