Python tf_logging.vlog方法代码示例

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _LogOpGradients(op, out_grads, in_grads):
  """Log the in and out grads of an op."""
  logging.vlog(1, "Gradient for '" + op.name + "'")

  def _FilterGrad(x):
    if x is None:
      return False
    if isinstance(x, (list, tuple)):
      return bool(x)
    else:
      return True

  logging.vlog(1, "  in  --> %s",
               ", ".join([x.name for x in out_grads if _FilterGrad(x)]))
  logging.vlog(1, "  out --> %s",
               ", ".join([x.name for x in in_grads if _FilterGrad(x)])) 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def register(self, candidate, name=None):
    """Registers a Python object "candidate" for the given "name".

    Args:
      candidate: The candidate object to add to the registry.
      name: An optional string specifying the registry key for the candidate.
            If None, candidate.__name__ will be used.
    Raises:
      KeyError: If same name is used twice.
    """
    if not name:
      name = candidate.__name__
    if name in self._registry:
      (filename, line_number, function_name, _) = (
          self._registry[name][_LOCATION_TAG])
      raise KeyError("Registering two %s with name '%s' !"
                     "(Previous registration was in %s %s:%d)" %
                     (self._name, name, function_name, filename, line_number))

    logging.vlog(1, "Registering %s (%s) in %s.", name, candidate, self._name)
    # stack trace is [this_function, Register(), user_function,...]
    # so the user function is #2.
    stack = traceback.extract_stack()
    self._registry[name] = {_TYPE_TAG: candidate, _LOCATION_TAG: stack[2]} 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _close_on_stop(self, sess, cancel_op, coord):
    """Close the queue when the Coordinator requests stop.

    Args:
      sess: A Session.
      cancel_op: The Operation to run.
      coord: Coordinator.
    """
    coord.register_thread(threading.current_thread())
    coord.wait_for_stop()
    try:
      sess.run(cancel_op)
    except Exception as e:
      # Intentionally ignore errors from cancel_op.
      logging.vlog(1, "Ignored exception: %s", str(e))
  # pylint: enable=broad-except 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _run(self, sess, enqueue_op, coord=None):

        if coord:
            coord.register_thread(threading.current_thread())
        decremented = False
        try:
            while True:
                if coord and coord.should_stop():
                    break
                try:
                    self.func(sess, enqueue_op)  # call enqueue function
                except self._queue_closed_exception_types:  # pylint: disable=catching-non-exception
                    # This exception indicates that a queue was closed.
                    with self._lock:
                        self._runs_per_session[sess] -= 1
                        decremented = True
                        if self._runs_per_session[sess] == 0:
                            try:
                                sess.run(self._close_op)
                            except Exception as e:
                                # Intentionally ignore errors from close_op.
                                logging.vlog(1, "Ignored exception: %s", str(e))
                        return
                except ValueError:  # ignore value error defined by queueing function
                    pass
        except Exception as e:
            # This catches all other exceptions.
            if coord:
                coord.request_stop(e)
            else:
                logging.error("Exception in QueueRunner: %s", str(e))
                with self._lock:
                    self._exceptions_raised.append(e)
                raise
        finally:
            # Make sure we account for all terminations: normal or errors.
            if not decremented:
                with self._lock:
                    self._runs_per_session[sess] -= 1 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _run(self, sess, enqueue_op, coord=None):

        if coord:
            coord.register_thread(threading.current_thread())
        decremented = False
        try:
            while True:
                if coord and coord.should_stop():
                    break
                try:
                    self.func(sess, enqueue_op)  # call enqueue function
                except self._queue_closed_exception_types:  # pylint: disable=catching-non-exception
                    # This exception indicates that a queue was closed.
                    with self._lock:
                        self._runs_per_session[sess] -= 1
                        decremented = True
                        if self._runs_per_session[sess] == 0:
                            try:
                                sess.run(self._close_op)
                            except Exception as e:
                                # Intentionally ignore errors from close_op.
                                logging.vlog(1, "Ignored exception: %s", str(e))
                        return
        except Exception as e:
            # This catches all other exceptions.
            if coord:
                coord.request_stop(e)
            else:
                logging.error("Exception in QueueRunner: %s", str(e))
                with self._lock:
                    self._exceptions_raised.append(e)
                raise
        finally:
            # Make sure we account for all terminations: normal or errors.
            if not decremented:
                with self._lock:
                    self._runs_per_session[sess] -= 1 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _close_on_stop(self, sess, cancel_op, coord):
    """Close the queue when the Coordinator requests stop.

    Args:
      sess: A Session.
      cancel_op: The Operation to run.
      coord: Coordinator.
    """
    coord.wait_for_stop()
    try:
      sess.run(cancel_op)
    except Exception as e:
      # Intentionally ignore errors from cancel_op.
      logging.vlog(1, "Ignored exception: %s", str(e))
  # pylint: enable=broad-except 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _run(self, sess, enqueue_op, feed_fn, coord=None):
    """Execute the enqueue op in a loop, close the queue in case of error.

    Args:
      sess: A `Session`.
      enqueue_op: The `Operation` to run.
      feed_fn: the feed function to pass to `sess.run`.
      coord: Optional `Coordinator` object for reporting errors and checking
        for stop conditions.

    """
    # TODO(jamieas): Reduce code duplication with `QueueRunner`.
    if coord:
      coord.register_thread(threading.current_thread())
    decremented = False
    try:
      while True:
        if coord and coord.should_stop():
          break
        try:
          feed_dict = None if feed_fn is None else feed_fn()
          sess.run(enqueue_op, feed_dict=feed_dict)
        except (errors.OutOfRangeError, errors.CancelledError):
          # This exception indicates that a queue was closed.
          with self._lock:
            self._runs_per_session[sess] -= 1
            decremented = True
            if self._runs_per_session[sess] == 0:
              try:
                sess.run(self._close_op)
              except Exception as e:
                # Intentionally ignore errors from close_op.
                logging.vlog(1, "Ignored exception: %s", str(e))
            return
    except Exception as e:
      # This catches all other exceptions.
      if coord:
        coord.request_stop(e)
      else:
        logging.error("Exception in QueueRunner: %s", str(e))
        with self._lock:
          self._exceptions_raised.append(e)
        raise
    finally:
      # Make sure we account for all terminations: normal or errors.
      if not decremented:
        with self._lock:
          self._runs_per_session[sess] -= 1 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _compute_numeric_jacobian(x, x_shape, x_data, y, y_shape, delta,
                              extra_feed_dict):
  """Computes the numeric Jacobian for dy/dx.

  Computes the numeric Jacobian by slightly perturbing the inputs and
  measuring the differences on the output.

  Args:
    x: the tensor "x".
    x_shape: the dimensions of x as a tuple or an array of ints.
    x_data: a numpy array as the input data for x
    y: the tensor "y".
    y_shape: the dimensions of y as a tuple or an array of ints.
    delta: the amount of perturbation we give to the input
    extra_feed_dict: dict that allows fixing specified tensor values
      during the jacobian calculation.

  Returns:
    A 2-d numpy array representing the Jacobian for dy/dx. It has "x_size" rows
    and "y_size" columns where "x_size" is the number of elements in x and
    "y_size" is the number of elements in y.
  """

  # To compute the jacobian, we treat x and y as one-dimensional vectors
  x_size = _product(x_shape) * (2 if x.dtype.is_complex else 1)
  y_size = _product(y_shape) * (2 if y.dtype.is_complex else 1)
  x_dtype = x.dtype.real_dtype.as_numpy_dtype
  y_dtype = y.dtype.real_dtype.as_numpy_dtype

  # Make sure we have the right types
  x_data = np.asarray(x_data, dtype=x.dtype.as_numpy_dtype)
  scale = np.asarray(2 * delta, dtype=y_dtype)[()]

  jacobian = np.zeros((x_size, y_size), dtype=x_dtype)
  # For each of the entry of x, we slightly perturbs this by adding and
  # subtracting a delta and then compute difference between the outputs. This
  # will give us one row of the Jacobian matrix.
  for row in range(x_size):
    x_pos = x_data.copy()
    x_neg = x_data.copy()
    x_pos.ravel().view(x_dtype)[row] += delta
    y_pos = y.eval(feed_dict=_extra_feeds(extra_feed_dict, {x: x_pos}))
    x_neg.ravel().view(x_dtype)[row] -= delta
    y_neg = y.eval(feed_dict=_extra_feeds(extra_feed_dict, {x: x_neg}))
    diff = (y_pos - y_neg) / scale
    jacobian[row, :] = diff.ravel().view(y_dtype)

  logging.vlog(1, "Numeric Jacobian =\n%s", jacobian)
  return jacobian 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _show_compute(self, show_dataflow):
    """Visualize the computation activity."""
    for dev_stats in self._step_stats.dev_stats:
      device_name = dev_stats.device
      device_pid = self._device_pids[device_name]
      is_gputrace = self._is_gputrace_device(device_name)

      for node_stats in dev_stats.node_stats:
        tid = node_stats.thread_id
        start_time = node_stats.all_start_micros
        end_time = node_stats.all_start_micros + node_stats.all_end_rel_micros
        self._emit_op(node_stats, device_pid, is_gputrace)

        if is_gputrace or node_stats.node_name == 'RecvTensor':
          continue

        _, _, inputs = self._parse_op_label(node_stats.timeline_label)
        for input_name in inputs:
          if input_name not in self._tensors:
            # This can happen when partitioning has inserted a Send/Recv.
            # We remove the numeric suffix so that the dataflow appears to
            # come from the original node.  Ideally, the StepStats would
            # contain logging for the Send and Recv nodes.
            index = input_name.rfind('/_')
            if index > 0:
              input_name = input_name[:index]

          if input_name in self._tensors:
            tensor = self._tensors[input_name]
            tensor.add_ref(start_time)
            tensor.add_unref(end_time - 1)

            if show_dataflow:
              # We use a different flow ID for every graph edge.
              create_time, create_pid, create_tid = self._flow_starts[
                  input_name]
              # Don't add flows when producer and consumer ops are on the same
              # pid/tid since the horizontal arrows clutter the visualization.
              if create_pid != device_pid or create_tid != tid:
                flow_id = self._alloc_flow_id()
                self._chrome_trace.emit_flow_start(input_name, create_time,
                                                   create_pid, create_tid,
                                                   flow_id)
                self._chrome_trace.emit_flow_end(input_name, start_time,
                                                 device_pid, tid, flow_id)
          else:
            logging.vlog(1, 'Can\'t find tensor %s - removed by CSE?',
                         input_name) 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _run(self, sess, enqueue_op, coord=None):
    """Execute the enqueue op in a loop, close the queue in case of error.

    Args:
      sess: A Session.
      enqueue_op: The Operation to run.
      coord: Optional Coordinator object for reporting errors and checking
        for stop conditions.
    """
    decremented = False
    try:
      # Make a cached callable from the `enqueue_op` to decrease the
      # Python overhead in the queue-runner loop.
      enqueue_callable = sess.make_callable(enqueue_op)
      while True:
        if coord and coord.should_stop():
          break
        try:
          enqueue_callable()
        except self._queue_closed_exception_types:  # pylint: disable=catching-non-exception
          # This exception indicates that a queue was closed.
          with self._lock:
            self._runs_per_session[sess] -= 1
            decremented = True
            if self._runs_per_session[sess] == 0:
              try:
                sess.run(self._close_op)
              except Exception as e:
                # Intentionally ignore errors from close_op.
                logging.vlog(1, "Ignored exception: %s", str(e))
            return
    except Exception as e:
      # This catches all other exceptions.
      if coord:
        coord.request_stop(e)
      else:
        logging.error("Exception in QueueRunner: %s", str(e))
        with self._lock:
          self._exceptions_raised.append(e)
        raise
    finally:
      # Make sure we account for all terminations: normal or errors.
      if not decremented:
        with self._lock:
          self._runs_per_session[sess] -= 1 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _run(self, sess, enqueue_op, coord=None):
    """Execute the enqueue op in a loop, close the queue in case of error.

    Args:
      sess: A Session.
      enqueue_op: The Operation to run.
      coord: Optional Coordinator object for reporting errors and checking
        for stop conditions.
    """
    decremented = False
    try:
      while True:
        if coord and coord.should_stop():
          break
        try:
          sess.run(enqueue_op)
        except self._queue_closed_exception_types:  # pylint: disable=catching-non-exception
          # This exception indicates that a queue was closed.
          with self._lock:
            self._runs_per_session[sess] -= 1
            decremented = True
            if self._runs_per_session[sess] == 0:
              try:
                sess.run(self._close_op)
              except Exception as e:
                # Intentionally ignore errors from close_op.
                logging.vlog(1, "Ignored exception: %s", str(e))
            return
    except Exception as e:
      # This catches all other exceptions.
      if coord:
        coord.request_stop(e)
      else:
        logging.error("Exception in QueueRunner: %s", str(e))
        with self._lock:
          self._exceptions_raised.append(e)
        raise
    finally:
      # Make sure we account for all terminations: normal or errors.
      if not decremented:
        with self._lock:
          self._runs_per_session[sess] -= 1 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _compute_numeric_jacobian(x, x_shape, x_data, y, y_shape, delta):
  """Computes the numeric Jacobian for dy/dx.

  Computes the numeric Jacobian by slightly perturbing the inputs and
  measuring the differences on the output.

  Args:
    x: the tensor "x".
    x_shape: the dimensions of x as a tuple or an array of ints.
    x_data: a numpy array as the input data for x
    y: the tensor "y".
    y_shape: the dimensions of y as a tuple or an array of ints.
    delta: the amount of perturbation we give to the input

  Returns:
    A 2-d numpy array representing the Jacobian for dy/dx. It has "x_size" rows
    and "y_size" columns where "x_size" is the number of elements in x and
    "y_size" is the number of elements in y.
  """

  # To compute the jacobian, we treat x and y as one-dimensional vectors
  x_size = _product(x_shape) * (2 if x.dtype.is_complex else 1)
  y_size = _product(y_shape) * (2 if y.dtype.is_complex else 1)
  x_dtype = x.dtype.real_dtype.as_numpy_dtype
  y_dtype = y.dtype.real_dtype.as_numpy_dtype

  # Make sure we have the right types
  x_data = np.asarray(x_data, dtype=x.dtype.as_numpy_dtype)
  scale = np.asarray(2 * delta, dtype=y_dtype)[()]

  jacobian = np.zeros((x_size, y_size), dtype=x_dtype)
  # For each of the entry of x, we slightly perturbs this by adding and
  # subtracting a delta and then compute difference between the outputs. This
  # will give us one row of the Jacobian matrix.
  for row in range(x_size):
    x_pos = x_data.copy()
    x_neg = x_data.copy()
    x_pos.ravel().view(x_dtype)[row] += delta
    y_pos = y.eval(feed_dict={x: x_pos})
    x_neg.ravel().view(x_dtype)[row] -= delta
    y_neg = y.eval(feed_dict={x: x_neg})
    diff = (y_pos - y_neg) / scale
    jacobian[row, :] = diff.ravel().view(y_dtype)

  logging.vlog(1, "Numeric Jacobian =\n%s", jacobian)
  return jacobian 

# 需要导入模块: from tensorflow.python.platform import tf_logging [as 别名]
# 或者: from tensorflow.python.platform.tf_logging import vlog [as 别名]
def _show_compute(self, show_dataflow):
    """Visualize the computation activity."""
    for dev_stats in self._step_stats.dev_stats:
      device_name = dev_stats.device
      device_pid = self._device_pids[device_name]
      is_gputrace = self._is_gputrace_device(device_name)

      for node_stats in dev_stats.node_stats:
        tid = node_stats.thread_id
        start_time = node_stats.all_start_micros
        end_time = node_stats.all_start_micros + node_stats.all_end_rel_micros
        self._emit_op(node_stats, device_pid, is_gputrace)

        if is_gputrace:
          continue

        _, _, inputs = self._parse_op_label(node_stats.timeline_label)
        for input_name in inputs:
          if input_name not in self._tensors:
            # This can happen when partitioning has inserted a Send/Recv.
            # We remove the numeric suffix so that the dataflow appears to
            # come from the original node.  Ideally, the StepStats would
            # contain logging for the Send and Recv nodes.
            index = input_name.rfind('/_')
            if index > 0:
              input_name = input_name[:index]

          if input_name in self._tensors:
            tensor = self._tensors[input_name]
            tensor.add_ref(start_time)
            tensor.add_unref(end_time - 1)

            if show_dataflow:
              # We use a different flow ID for every graph edge.
              create_time, create_pid, create_tid = self._flow_starts[
                  input_name]
              # Don't add flows when producer and consumer ops are on the same
              # pid/tid since the horizontal arrows clutter the visualization.
              if create_pid != device_pid or create_tid != tid:
                flow_id = self._alloc_flow_id()
                self._chrome_trace.emit_flow_start(input_name, create_time,
                                                   create_pid, create_tid,
                                                   flow_id)
                self._chrome_trace.emit_flow_end(input_name, start_time,
                                                 device_pid, tid, flow_id)
          else:
            logging.vlog(1, 'Can\'t find tensor %s - removed by CSE?',
                         input_name)