Python tensorflow.python.training.training_util.global_step() Examples

def summary_computed(self, sess, summary, global_step=None):
    """Indicate that a summary was computed.

    Args:
      sess: A `Session` object.
      summary: A Summary proto, or a string holding a serialized summary proto.
      global_step: Int. global step this summary is associated with. If `None`,
        it will try to fetch the current step.

    Raises:
      TypeError: if 'summary' is not a Summary proto or a string.
      RuntimeError: if the Supervisor was created without a `logdir`.
    """
    if not self._summary_writer:
      raise RuntimeError("Writing a summary requires a summary writer.")
    if global_step is None and self.global_step is not None:
      global_step = training_util.global_step(sess, self.global_step)
    self._summary_writer.add_summary(summary, global_step) 

def __init__(self, sv, sess, step_counter=None):
    """Create a `SVStepCounterThread`.

    Args:
      sv: A `Supervisor`.
      sess: A `Session`.
      step_counter: A `Tensor` holding the step counter. By defaults, it uses
        sv.global_step.
    """
    super(SVStepCounterThread, self).__init__(sv.coord, sv.save_summaries_secs)
    self._sv = sv
    self._sess = sess
    self._last_time = 0.0
    self._last_step = 0
    step_counter = sv.global_step if step_counter is None else step_counter
    self._step_counter = step_counter
    self._summary_tag = "%s/sec" % self._step_counter.op.name 

def run_loop(self):
    # Count the steps.
    current_step = training_util.global_step(self._sess, self._sv.global_step)
    added_steps = current_step - self._last_step
    self._last_step = current_step
    # Measure the elapsed time.
    current_time = time.time()
    elapsed_time = current_time - self._last_time
    self._last_time = current_time
    # Reports the number of steps done per second
    steps_per_sec = added_steps / elapsed_time
    summary = Summary(value=[Summary.Value(tag=self._summary_tag,
                                           simple_value=steps_per_sec)])
    if self._sv.summary_writer:
      self._sv.summary_writer.add_summary(summary, current_step)
    logging.log_first_n(logging.INFO, "%s: %g", 10,
                        self._summary_tag, steps_per_sec) 

def summary_computed(self, sess, summary, global_step=None):
    """Indicate that a summary was computed.

    Args:
      sess: A `Session` object.
      summary: A Summary proto, or a string holding a serialized summary proto.
      global_step: Int. global step this summary is associated with. If `None`,
        it will try to fetch the current step.

    Raises:
      TypeError: if 'summary' is not a Summary proto or a string.
      RuntimeError: if the Supervisor was created without a `logdir`.
    """
    if not self._summary_writer:
      raise RuntimeError("Writing a summary requires a summary writer.")
    if global_step is None and self.global_step is not None:
      global_step = training_util.global_step(sess, self.global_step)
    self._summary_writer.add_summary(summary, global_step) 

def summary_computed(self, sess, summary, global_step=None):
    """Indicate that a summary was computed.

    Args:
      sess: A `Session` object.
      summary: A Summary proto, or a string holding a serialized summary proto.
      global_step: Int. global step this summary is associated with. If `None`,
        it will try to fetch the current step.

    Raises:
      TypeError: if 'summary' is not a Summary proto or a string.
      RuntimeError: if the Supervisor was created without a `logdir`.
    """
    if not self._summary_writer:
      raise RuntimeError("Writing a summary requires a summary writer.")
    if global_step is None and self.global_step is not None:
      global_step = training_util.global_step(sess, self.global_step)
    self._summary_writer.add_summary(summary, global_step) 

def __init__(self, sv, sess, step_counter=None):
    """Create a `SVStepCounterThread`.

    Args:
      sv: A `Supervisor`.
      sess: A `Session`.
      step_counter: A `Tensor` holding the step counter. By defaults, it uses
        sv.global_step.
    """
    super(SVStepCounterThread, self).__init__(sv.coord, sv.save_summaries_secs)
    self._sv = sv
    self._sess = sess
    self._last_time = 0.0
    self._last_step = 0
    step_counter = sv.global_step if step_counter is None else step_counter
    self._step_counter = step_counter
    self._summary_tag = "%s/sec" % self._step_counter.op.name 

def summary_computed(self, sess, summary, global_step=None):
    """Indicate that a summary was computed.

    Args:
      sess: A `Session` object.
      summary: A Summary proto, or a string holding a serialized summary proto.
      global_step: Int. global step this summary is associated with. If `None`,
        it will try to fetch the current step.

    Raises:
      TypeError: if 'summary' is not a Summary proto or a string.
      RuntimeError: if the Supervisor was created without a `logdir`.
    """
    if not self._summary_writer:
      raise RuntimeError("Writing a summary requires a summary writer.")
    if global_step is None and self.global_step is not None:
      global_step = training_util.global_step(sess, self.global_step)
    self._summary_writer.add_summary(summary, global_step) 

def run_loop(self):
    # Count the steps.
    current_step = training_util.global_step(self._sess, self._step_counter)
    added_steps = current_step - self._last_step
    self._last_step = current_step
    # Measure the elapsed time.
    current_time = time.time()
    elapsed_time = current_time - self._last_time
    self._last_time = current_time
    # Reports the number of steps done per second
    if elapsed_time > 0.:
      steps_per_sec = added_steps / elapsed_time
    else:
      steps_per_sec = float("inf")
    summary = Summary(value=[Summary.Value(tag=self._summary_tag,
                                           simple_value=steps_per_sec)])
    if self._sv.summary_writer:
      self._sv.summary_writer.add_summary(summary, current_step)
    logging.log_first_n(logging.INFO, "%s: %g", 10,
                        self._summary_tag, steps_per_sec) 

def run_loop(self):
    # Count the steps.
    current_step = training_util.global_step(self._sess, self._step_counter)
    added_steps = current_step - self._last_step
    self._last_step = current_step
    # Measure the elapsed time.
    current_time = time.time()
    elapsed_time = current_time - self._last_time
    self._last_time = current_time
    # Reports the number of steps done per second
    if elapsed_time > 0.:
      steps_per_sec = added_steps / elapsed_time
    else:
      steps_per_sec = float("inf")
    summary = Summary(value=[Summary.Value(tag=self._summary_tag,
                                           simple_value=steps_per_sec)])
    if self._sv.summary_writer:
      self._sv.summary_writer.add_summary(summary, current_step)
    logging.log_first_n(logging.INFO, "%s: %g", 10,
                        self._summary_tag, steps_per_sec) 

def summary_computed(self, sess, summary, global_step=None):
    """Indicate that a summary was computed.

    Args:
      sess: A `Session` object.
      summary: A Summary proto, or a string holding a serialized summary proto.
      global_step: Int. global step this summary is associated with. If `None`,
        it will try to fetch the current step.

    Raises:
      TypeError: if 'summary' is not a Summary proto or a string.
      RuntimeError: if the Supervisor was created without a `logdir`.
    """
    if not self._summary_writer:
      raise RuntimeError("Writing a summary requires a summary writer.")
    if global_step is None and self.global_step is not None:
      global_step = training_util.global_step(sess, self.global_step)
    self._summary_writer.add_summary(summary, global_step) 

def run_loop(self):
    # Count the steps.
    current_step = training_util.global_step(self._sess, self._step_counter)
    added_steps = current_step - self._last_step
    self._last_step = current_step
    # Measure the elapsed time.
    current_time = time.time()
    elapsed_time = current_time - self._last_time
    self._last_time = current_time
    # Reports the number of steps done per second
    if elapsed_time > 0.:
      steps_per_sec = added_steps / elapsed_time
    else:
      steps_per_sec = float("inf")
    summary = Summary(value=[Summary.Value(tag=self._summary_tag,
                                           simple_value=steps_per_sec)])
    if self._sv.summary_writer:
      self._sv.summary_writer.add_summary(summary, current_step)
    logging.log_first_n(logging.INFO, "%s: %g", 10,
                        self._summary_tag, steps_per_sec) 

def run_loop(self):
    # Count the steps.
    current_step = training_util.global_step(self._sess, self._sv.global_step)
    added_steps = current_step - self._last_step
    self._last_step = current_step
    # Measure the elapsed time.
    current_time = time.time()
    elapsed_time = current_time - self._last_time
    self._last_time = current_time
    # Reports the number of steps done per second
    steps_per_sec = added_steps / elapsed_time
    summary = Summary(value=[Summary.Value(tag=self._summary_tag,
                                           simple_value=steps_per_sec)])
    if self._sv.summary_writer:
      self._sv.summary_writer.add_summary(summary, current_step)
    logging.log_first_n(logging.INFO, "%s: %g", 10,
                        self._summary_tag, steps_per_sec) 

def summary_computed(self, sess, summary, global_step=None):
    """Indicate that a summary was computed.

    Args:
      sess: A `Session` object.
      summary: A Summary proto, or a string holding a serialized summary proto.
      global_step: Int. global step this summary is associated with. If `None`,
        it will try to fetch the current step.

    Raises:
      TypeError: if 'summary' is not a Summary proto or a string.
      RuntimeError: if the Supervisor was created without a `logdir`.
    """
    if not self._summary_writer:
      raise RuntimeError("Writing a summary requires a summary writer.")
    if global_step is None and self.global_step is not None:
      global_step = training_util.global_step(sess, self.global_step)
    self._summary_writer.add_summary(summary, global_step) 

def run_loop(self):
    # Count the steps.
    current_step = training_util.global_step(self._sess, self._sv.global_step)
    added_steps = current_step - self._last_step
    self._last_step = current_step
    # Measure the elapsed time.
    current_time = time.time()
    elapsed_time = current_time - self._last_time
    self._last_time = current_time
    # Reports the number of steps done per second
    steps_per_sec = added_steps / elapsed_time
    summary = Summary(value=[Summary.Value(tag=self._summary_tag,
                                           simple_value=steps_per_sec)])
    if self._sv.summary_writer:
      self._sv.summary_writer.add_summary(summary, current_step)
    logging.log_first_n(logging.INFO, "%s: %g", 10,
                        self._summary_tag, steps_per_sec) 

def __init__(self, sv, sess, step_counter=None):
    """Create a `SVStepCounterThread`.

    Args:
      sv: A `Supervisor`.
      sess: A `Session`.
      step_counter: A `Tensor` holding the step counter. By defaults, it uses
        sv.global_step.
    """
    super(SVStepCounterThread, self).__init__(sv.coord, sv.save_summaries_secs)
    self._sv = sv
    self._sess = sess
    self._last_time = 0.0
    self._last_step = 0
    step_counter = sv.global_step if step_counter is None else step_counter
    self._step_counter = step_counter
    self._summary_tag = "%s/sec" % self._step_counter.op.name 

def _default_global_step_tensor(self):
    """Returns the global_step from the default graph.

    Returns:
      The global step `Tensor` or `None`.
    """
    try:
      gs = ops.get_default_graph().get_tensor_by_name("global_step:0")
      if gs.dtype.base_dtype in [dtypes.int32, dtypes.int64]:
        return gs
      else:
        logging.warning("Found 'global_step' is not an int type: %s", gs.dtype)
        return None
    except KeyError:
      return None 

def start_loop(self):
    self._last_time = time.time()
    self._last_step = training_util.global_step(
        self._sess, self._sv.global_step) 

def _wait_for_step(sess, global_step, step):
  """Wait till the global step has reached at least 'step'.

  Args:
    sess: A session.
    global_step: A Tensor.
    step: Int.  The global step to reach.
  """
  while True:
    if training_util.global_step(sess, global_step) >= step:
      break
    time.sleep(1.0) 

def __init__(self, sv, sess):
    """Create a `SVStepCounterThread`.

    Args:
      sv: A `Supervisor`.
      sess: A `Session`.
    """
    super(SVStepCounterThread, self).__init__(sv.coord, sv.save_summaries_secs)
    self._sv = sv
    self._sess = sess
    self._last_time = 0.0
    self._last_step = 0
    self._summary_tag = "%s/sec" % self._sv.global_step.op.name 

def global_step(self):
    """Return the global_step Tensor used by the supervisor.

    Returns:
      An integer Tensor for the global_step.
    """
    return self._global_step 

def run_loop(self):
    self._sv.saver.save(self._sess, self._sv.save_path,
                        global_step=self._sv.global_step)
    if self._sv.summary_writer and self._sv.global_step is not None:
      current_step = training_util.global_step(self._sess, self._sv.global_step)
      self._sv.summary_writer.add_session_log(
          SessionLog(status=SessionLog.CHECKPOINT,
                     checkpoint_path=self._sv.save_path),
          current_step)


# TODO(sherrym): All non-PEP8 compliant names will be deprecated shortly. 

def _init_global_step(self, global_step=USE_DEFAULT):
    """Initializes global_step.

    Args:
      global_step: An integer Tensor of size 1 that counts steps. If
        set to USE_DEFAULT, creates global_step tensor.
    """
    if global_step is Supervisor.USE_DEFAULT:
      global_step = self._get_first_op_from_collection(
          ops.GraphKeys.GLOBAL_STEP)
      if global_step is None:
        global_step = self._default_global_step_tensor()
        if global_step is not None:
          ops.add_to_collection(ops.GraphKeys.GLOBAL_STEP, global_step)
    self._global_step = global_step 

def _init_global_step(self, global_step=USE_DEFAULT):
    """Initializes global_step.

    Args:
      global_step: An integer Tensor of size 1 that counts steps. If
        set to USE_DEFAULT, creates global_step tensor.
    """
    if global_step is Supervisor.USE_DEFAULT:
      global_step = self._get_first_op_from_collection(
          ops.GraphKeys.GLOBAL_STEP)
      if global_step is None:
        global_step = self._default_global_step_tensor()
        if global_step is not None:
          ops.add_to_collection(ops.GraphKeys.GLOBAL_STEP, global_step)
    self._global_step = global_step 

def _wait_for_step(sess, global_step, step):
  """Wait till the global step has reached at least 'step'.

  Args:
    sess: A session.
    global_step: A Tensor.
    step: Int.  The global step to reach.
  """
  while True:
    if training_util.global_step(sess, global_step) >= step:
      break
    time.sleep(1.0) 

def run_loop(self):
    self._sv.saver.save(self._sess, self._sv.save_path,
                        global_step=self._sv.global_step)
    if self._sv.summary_writer and self._sv.global_step is not None:
      current_step = training_util.global_step(self._sess, self._sv.global_step)
      self._sv.summary_writer.add_session_log(
          SessionLog(status=SessionLog.CHECKPOINT,
                     checkpoint_path=self._sv.save_path),
          current_step)


# TODO(sherrym): All non-PEP8 compliant names will be deprecated shortly. 

def start_loop(self):
    self._last_time = time.time()
    self._last_step = training_util.global_step(
        self._sess, self._sv.global_step) 

def __init__(self, sv, sess):
    """Create a `SVStepCounterThread`.

    Args:
      sv: A `Supervisor`.
      sess: A `Session`.
    """
    super(SVStepCounterThread, self).__init__(sv.coord, sv.save_summaries_secs)
    self._sv = sv
    self._sess = sess
    self._last_time = 0.0
    self._last_step = 0
    self._summary_tag = "%s/sec" % self._sv.global_step.op.name 

def _default_global_step_tensor(self):
    """Returns the global_step from the default graph.

    Returns:
      The global step `Tensor` or `None`.
    """
    try:
      gs = ops.get_default_graph().get_tensor_by_name("global_step:0")
      if gs.dtype.base_dtype in [dtypes.int32, dtypes.int64]:
        return gs
      else:
        logging.warning("Found 'global_step' is not an int type: %s", gs.dtype)
        return None
    except KeyError:
      return None 

def global_step(self):
    """Return the global_step Tensor used by the supervisor.

    Returns:
      An integer Tensor for the global_step.
    """
    return self._global_step 

def _wait_for_step(sess, global_step, step):
  """Wait till the global step has reached at least 'step'.

  Args:
    sess: A session.
    global_step: A Tensor.
    step: Int.  The global step to reach.
  """
  while True:
    if training_util.global_step(sess, global_step) >= step:
      break
    time.sleep(1.0)