Tensorflow抱怨在圖形還原期間丟失了feed_dict

Question

我已經爲圖像分類構建了CNN。在訓練期間，我保存了幾個檢查點。數據通過feed_dictionary饋入網絡。

現在我想恢復失敗的模型，我不知道爲什麼。有代碼的重要線路如下：

with tf.Graph().as_default(): 

.... 

if checkpoint_dir is not None: 
    checkpoint_saver = tf.train.Saver() 
    session_hooks.append(tf.train.CheckpointSaverHook(checkpoint_dir, 
                 save_secs=flags.save_interval_secs, 
                 saver=checkpoint_saver)) 
.... 

with tf.train.MonitoredTrainingSession(
     save_summaries_steps=flags.save_summaries_steps, 
     hooks=session_hooks, 
     config=tf.ConfigProto(
      log_device_placement=flags.log_device_placement)) as mon_sess: 

    checkpoint = tf.train.get_checkpoint_state(checkpoint_dir) 
    if checkpoint and checkpoint.model_checkpoint_path: 

     # restoring from the checkpoint file 
     checkpoint_saver.restore(mon_sess, checkpoint.model_checkpoint_path) 

     global_step_restore = checkpoint.model_checkpoint_path.split('/')[-1].split('-')[-1] 
     print("Model restored from checkpoint: global_step = %s" % global_step_restore) 

行 「checkpoint_saver.restore」 拋出一個錯誤：

回溯（最近通話最後一個）： 文件「C：\ Program Files文件\ Anaconda3 \ （* args） 文件「C：\ Program Files \ Anaconda3 \ envs \ tensorflow \ lib \」文件名爲「envs \ tensorflow \ lib \ site-packages \ tensorflow \ python \ client \ session.py」，第1022行，在_do_call中 site_packages \ tensorflow \ python \ client \ session.py「，第1004行，在_run_fn status，run_metadata） 文件」C：\ Program Files \ Anaconda3 \ envs \ tensorflow \ lib \ contextlib.py「，第6行6，在退出 next（self.gen） 在raise_exception_on_not_ok_status文件「C：\ Program Files \ Anaconda3 \ envs \ tensorflow \ lib \ site-packages \ tensorflow \ python \ framework \ errors_impl.py」行469 pywrap_tensorflow.TF_GetCode（status）） tensorflow.python.framework.errors_impl.InvalidArgumentError：您必須爲dtype float提供佔位符張量'input_images'的值 [[Node：input_images = Placeholderdtype = DT_FLOAT，shape = []，_device =「/ job：localhost/replica：0/task：0/cpu：0」]]

任何知道如何解決這個問題？爲什麼我只需要填充的feed_dictionary來恢復圖形？

在此先感謝！

更新：

這是保護對象的恢復方法的代碼：

def restore(self, sess, save_path): 
    """Restores previously saved variables. 

    This method runs the ops added by the constructor for restoring variables. 
    It requires a session in which the graph was launched. The variables to 
    restore do not have to have been initialized, as restoring is itself a way 
    to initialize variables. 

    The `save_path` argument is typically a value previously returned from a 
    `save()` call, or a call to `latest_checkpoint()`. 

    Args: 
     sess: A `Session` to use to restore the parameters. 
     save_path: Path where parameters were previously saved. 
    """ 
    if self._is_empty: 
     return 
    sess.run(self.saver_def.restore_op_name, 
      {self.saver_def.filename_tensor_name: save_path}) 

什麼我不明白：爲什麼圖表立即執行？我使用錯誤的方法嗎？我只想恢復所有可訓練的變數。

Answer 1

問題是用於進程日誌原因由SessionRunHook：

原始鉤：

class _LoggerHook(tf.train.SessionRunHook): 
    """Logs loss and runtime.""" 

    def begin(self): 
    self._step = -1 

    def before_run(self, run_context): 
    self._step += 1 
    self._start_time = time.time() 
    return tf.train.SessionRunArgs(loss) # Asks for loss value. 

    def after_run(self, run_context, run_values): 
    duration = time.time() - self._start_time 
    loss_value = run_values.results 
    if self._step % 5 == 0: 
     num_examples_per_step = FLAGS.batch_size 
     examples_per_sec = num_examples_per_step/duration 
     sec_per_batch = float(duration) 

     format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f ' 
        'sec/batch)') 
     print (format_str % (datetime.now(), self._step, loss_value, 
          examples_per_sec, sec_per_batch)) 

改性鉤：

class _LoggerHook(tf.train.SessionRunHook): 
    """Logs loss and runtime.""" 

    def __init__(self, flags, loss_op): 
     self._flags = flags 
     self._loss_op = loss_op 
     self._start_time = time.time() 

    def begin(self): 
     self._step = 0 

    def before_run(self, run_context): 
     if self._step == 0: 
      run_args = None 
     else: 
      run_args = tf.train.SessionRunArgs(self._loss_op) 

     return run_args 

    def after_run(self, run_context, run_values): 

     if self._step > 0: 
      duration_n_steps = time.time() - self._start_time 
      loss_value = run_values.results 
      if self._step % self._flags.log_every_n_steps == 0: 
       num_examples_per_step = self._flags.batch_size 

       duration = duration_n_steps/self._flags.log_every_n_steps 
       examples_per_sec = num_examples_per_step/duration 
       sec_per_batch = float(duration) 

       format_str = ('%s: step %d, loss = %.2f (%.1f examples/sec; %.3f ' 
           'sec/batch)') 
       print(format_str % (datetime.now(), self._step, loss_value, 
            examples_per_sec, sec_per_batch)) 

       self._start_time = time.time() 
     self._step += 1 

說明：

測井現在skiped第一次迭代。因此，由Saver.restore（..）執行的session.run不再需要填充的飼料字典。