Python summary.scalar函数代码示例

  def apply_gradients(self, grads_and_vars, global_step=None, name=None):
    gradients = []
    # Number of stale gradients.
    stale_counter = variable_scope.get_variable(
        "stale_counter", [],
        initializer=init_ops.zeros_initializer(),
        trainable=False)

    def _AcceptGradientOp():
      with ops.control_dependencies(
          [self._opt.apply_gradients(
              grads_and_vars, global_step=global_step, name=name)]):
        return gen_array_ops.identity(0.0)

    def _DropGradientOp():
      return gen_array_ops.identity(1.0)

    for grad_and_var in grads_and_vars:
      grad = grad_and_var[0]
      if isinstance(grad, ops.Tensor):
        gradients.append(grad)
      else:
        gradients.append(grad.op)

    with ops.control_dependencies(gradients), ops.colocate_with(global_step):
      staleness = gen_array_ops.reshape(
          global_step - self._local_step, shape=())
      conditional_update = stale_counter.assign_add(control_flow_ops.cond(
          gen_math_ops.less_equal(staleness, self._staleness),
          _AcceptGradientOp, _DropGradientOp))

    summary.scalar(
        "Gradient staleness percentage",
        stale_counter / (math_ops.cast(global_step + 1, dtypes.float32)))
    return conditional_update

  def linear_logit_fn(features):
    """Linear model logit_fn.

    Args:
      features: This is the first item returned from the `input_fn`
                passed to `train`, `evaluate`, and `predict`. This should be a
                single `Tensor` or `dict` of same.

    Returns:
      A `Tensor` representing the logits.
    """
    cols_to_vars = {}
    logits = feature_column_lib.linear_model(
        features=features,
        feature_columns=feature_columns,
        units=units,
        sparse_combiner=sparse_combiner,
        cols_to_vars=cols_to_vars)
    bias = cols_to_vars.pop('bias')
    if units > 1:
      summary.histogram('bias', bias)
    else:
      # If units == 1, the bias value is a length-1 list of a scalar Tensor,
      # so we should provide a scalar summary.
      summary.scalar('bias', bias[0][0])
    summary.scalar('fraction_of_zero_weights',
                   _compute_fraction_of_zero(cols_to_vars))
    return logits

 def test_train_worker_monitor(self):
   # We need to explicitly set device due to check on non-chief workers
   # requiring all variables to have a device assigned.
   with ops.Graph().as_default() as g, g.device('/cpu:0'):
     global_step = variables_lib.create_global_step(g)
     train_op = state_ops.assign_add(global_step, 1)
     loss_op = constant_op.constant(2.0)
     summary.scalar('loss', loss_op)
     # Add explicit "local" init op to initialize all variables
     # as there's no chief to init here.
     init_op = variables.global_variables_initializer()
     ops.add_to_collection(ops.GraphKeys.LOCAL_INIT_OP, init_op)
     # Create worker monitors where one should be active on the worker
     # and the other chief exclusive.
     chief_exclusive_monitor = _BaseMonitorWrapper(False)
     all_workers_monitor = _BaseMonitorWrapper(True)
     with self.test_session(g):
       loss = learn.graph_actions.train(
           g,
           output_dir=self._output_dir,
           global_step_tensor=global_step,
           train_op=train_op,
           loss_op=loss_op,
           supervisor_is_chief=False,
           steps=1,
           monitors=[chief_exclusive_monitor, all_workers_monitor])
     self.assertEqual(2.0, loss)
     self.assertTrue(not chief_exclusive_monitor.is_active and
                     all_workers_monitor.is_active,
                     'Only non-chief runnable monitor must have been active.')
     self.assertTrue(not chief_exclusive_monitor.has_step and
                     all_workers_monitor.has_step,
                     'Only non-chief runnable monitor must have a step.')

 def test_train_summaries(self):
   with ops.Graph().as_default() as g, self.test_session(g):
     with ops.control_dependencies(self._build_inference_graph()):
       train_op = state_ops.assign_add(variables_lib.get_global_step(), 1)
     loss_op = constant_op.constant(2.0)
     summary.scalar('loss', loss_op)
     self._assert_summaries(self._output_dir)
     self._assert_ckpt(self._output_dir, False)
     loss = learn.graph_actions.train(
         g,
         output_dir=self._output_dir,
         train_op=train_op,
         loss_op=loss_op,
         steps=1)
     # TODO(ebrevdo,ptucker,ispir): this meta_graph_def lacks the
     # SaverDef, so we can't add it to the summary assertion test below.
     # meta_graph_def = meta_graph.create_meta_graph_def()
     self.assertEqual(2.0, loss)
     self._assert_summaries(
         self._output_dir,
         expected_graphs=[g],
         expected_summaries={1: {
             'loss': 2.0
         }})
     self._assert_ckpt(self._output_dir, True)

def prefetch_queue(tensors,
                   capacity=8,
                   num_threads=1,
                   dynamic_pad=False,
                   shared_name=None,
                   name=None):
  """Creates a queue to prefetch tensors from `tensors`.

  A queue runner for enqueuing tensors into the prefetch_queue is automatically
  added to the TF QueueRunners collection.

  Example:
  This is for example useful to pre-assemble input batches read with
  `tf.train.batch()` and enqueue the pre-assembled batches.  Ops that dequeue
  from the pre-assembled queue will not pay the cost of assembling the batch.

  images, labels = tf.train.batch([image, label], batch_size=32, num_threads=4)
  batch_queue = prefetch_queue([images, labels])
  images, labels = batch_queue.dequeue()
  logits = Net(images)
  loss = Loss(logits, labels)

  Args:
    tensors: A list or dictionary of `Tensors` to enqueue in the buffer.
    capacity: An integer. The maximum number of elements in the queue.
    num_threads: An integer.  Number of threads running the enqueue op.
    dynamic_pad: Boolean.  Whether to allow variable dimensions in input shapes.
    shared_name: (optional). If set, this queue will be shared under the given
      name across multiple sessions.
    name: (Optional) A name for the operations.

  Returns:
    A queue from which you can dequeue tensors with the same type and shape
    as `tensors`.
  """
  if isinstance(tensors, dict):
    # Need to wrap the keys and values in list() since Python3 returns views.
    # We sort the keys so the order is consistent across runs.
    names = list(sorted(tensors.keys()))
    tensor_list = list([tensors[n] for n in names])
  else:
    names = None
    tensor_list = tensors

  with ops.name_scope(name, "prefetch_queue", tensor_list) as name:
    dtypes = [t.dtype for t in tensor_list]
    shapes = [t.get_shape() for t in tensor_list]
    queue = _which_queue(dynamic_pad)(
        capacity=capacity,
        dtypes=dtypes,
        shapes=shapes,
        names=names,
        shared_name=shared_name)
    enqueue_op = queue.enqueue(tensors)
    queue_runner.add_queue_runner(
        queue_runner.QueueRunner(queue, [enqueue_op] * num_threads))
    summary.scalar(
        "fraction_of_%d_full" % capacity,
        math_ops.cast(queue.size(), _dtypes.float32) * (1. / capacity))
    return queue

    def _ModelFn(features, labels, mode):
      if is_training:
        logits_out = self._BuildGraph(features)
      else:
        graph_def = self._GetGraphDef(use_trt, batch_size, model_dir)
        logits_out = importer.import_graph_def(
            graph_def,
            input_map={INPUT_NODE_NAME: features},
            return_elements=[OUTPUT_NODE_NAME + ':0'],
            name='')[0]

      loss = losses.sparse_softmax_cross_entropy(
          labels=labels, logits=logits_out)
      summary.scalar('loss', loss)

      classes_out = math_ops.argmax(logits_out, axis=1, name='classes_out')
      accuracy = metrics.accuracy(
          labels=labels, predictions=classes_out, name='acc_op')
      summary.scalar('accuracy', accuracy[1])

      if mode == ModeKeys.EVAL:
        return EstimatorSpec(
            mode, loss=loss, eval_metric_ops={'accuracy': accuracy})
      elif mode == ModeKeys.TRAIN:
        optimizer = AdamOptimizer(learning_rate=1e-2)
        train_op = optimizer.minimize(loss, global_step=get_global_step())
        return EstimatorSpec(mode, loss=loss, train_op=train_op)

  def gradient_clipping(grads_and_vars):
    """Internal function for adaptive clipping."""
    grads, variables = zip(*grads_and_vars)

    norm = clip_ops.global_norm(grads)

    max_norm, log_mean = _adaptive_max_norm(norm, std_factor, decay,
                                            global_step, epsilon, name)

    # reports the max gradient norm for debugging
    if report_summary:
      summary.scalar("global_norm/adaptive_max_gradient_norm", max_norm)

    # factor will be 1. if norm is smaller than max_norm
    factor = array_ops.where(norm < max_norm,
                             array_ops.ones_like(norm),
                             math_ops.exp(log_mean) / norm)

    if static_max_norm is not None:
      factor = math_ops.minimum(static_max_norm / norm, factor)

    # apply factor
    clipped_grads = []
    for grad in grads:
      if grad is None:
        clipped_grads.append(None)
      elif isinstance(grad, ops.IndexedSlices):
        clipped_grads.append(
            ops.IndexedSlices(grad.values * factor, grad.indices,
                              grad.dense_shape))
      else:
        clipped_grads.append(grad * factor)

    return list(zip(clipped_grads, variables))

  def testSummariesAreFlushedToDisk(self):
    checkpoint_dir = os.path.join(self.get_temp_dir(), 'summaries_are_flushed')
    logdir = os.path.join(self.get_temp_dir(), 'summaries_are_flushed_eval')
    if gfile.Exists(logdir):
      gfile.DeleteRecursively(logdir)

    # Train a Model to completion:
    self._train_model(checkpoint_dir, num_steps=300)

    # Create the model (which can be restored).
    inputs = constant_op.constant(self._inputs, dtype=dtypes.float32)
    logistic_classifier(inputs)

    names_to_values = {'bread': 3.4, 'cheese': 4.5, 'tomato': 2.0}

    for k in names_to_values:
      v = names_to_values[k]
      summary_lib.scalar(k, v)

    evaluation.evaluate_repeatedly(
        checkpoint_dir=checkpoint_dir,
        hooks=[evaluation.SummaryAtEndHook(log_dir=logdir),],
        max_number_of_evaluations=1)

    self._verify_summaries(logdir, names_to_values)

  def testSummaryWithFamilyMetaGraphExport(self):
    with ops.name_scope('outer'):
      i = constant_op.constant(11)
      summ = summary_lib.scalar('inner', i)
      self.assertEquals(summ.op.name, 'outer/inner')
      summ_f = summary_lib.scalar('inner', i, family='family')
      self.assertEquals(summ_f.op.name, 'outer/family/inner')

    metagraph_def, _ = meta_graph.export_scoped_meta_graph(export_scope='outer')

    with ops.Graph().as_default() as g:
      meta_graph.import_scoped_meta_graph(metagraph_def, graph=g,
                                          import_scope='new_outer')
      # The summaries should exist, but with outer scope renamed.
      new_summ = g.get_tensor_by_name('new_outer/inner:0')
      new_summ_f = g.get_tensor_by_name('new_outer/family/inner:0')

      # However, the tags are unaffected.
      with self.cached_session() as s:
        new_summ_str, new_summ_f_str = s.run([new_summ, new_summ_f])
        new_summ_pb = summary_pb2.Summary()
        new_summ_pb.ParseFromString(new_summ_str)
        self.assertEquals('outer/inner', new_summ_pb.value[0].tag)
        new_summ_f_pb = summary_pb2.Summary()
        new_summ_f_pb.ParseFromString(new_summ_f_str)
        self.assertEquals('family/outer/family/inner',
                          new_summ_f_pb.value[0].tag)

  def create_estimator_spec(
      self, features, mode, logits, labels=None, train_op_fn=None):
    """See `Head`."""
    with ops.name_scope('head'):
      logits = head_lib._check_logits(logits, self.logits_dimension)  # pylint:disable=protected-access

      # Predict.
      pred_keys = prediction_keys.PredictionKeys
      with ops.name_scope(None, 'predictions', (logits,)):
        probabilities = math_ops.sigmoid(logits, name=pred_keys.PROBABILITIES)
        predictions = {
            pred_keys.LOGITS: logits,
            pred_keys.PROBABILITIES: probabilities,
        }
      if mode == model_fn.ModeKeys.PREDICT:
        return model_fn.EstimatorSpec(
            mode=model_fn.ModeKeys.PREDICT,
            predictions=predictions,
            export_outputs={
                '': export_output.ClassificationOutput(scores=probabilities)
            })

      # Eval.
      unweighted_loss, processed_labels = self.create_loss(
          features=features, mode=mode, logits=logits, labels=labels)
      # Averages loss over classes.
      per_example_loss = math_ops.reduce_mean(
          unweighted_loss, axis=-1, keep_dims=True)
      weights = head_lib._weights(features, self._weight_column)  # pylint:disable=protected-access
      training_loss = losses.compute_weighted_loss(
          per_example_loss, weights=weights, reduction=losses.Reduction.SUM)
      if mode == model_fn.ModeKeys.EVAL:
        return model_fn.EstimatorSpec(
            mode=model_fn.ModeKeys.EVAL,
            predictions=predictions,
            loss=training_loss,
            eval_metric_ops=self._eval_metric_ops(
                labels=processed_labels,
                probabilities=probabilities,
                weights=weights,
                per_example_loss=per_example_loss))

      # Train.
      if train_op_fn is None:
        raise ValueError('train_op_fn can not be None.')
    with ops.name_scope(''):
      summary.scalar(
          head_lib._summary_key(self._name, metric_keys.MetricKeys.LOSS),  # pylint:disable=protected-access
          training_loss)
      summary.scalar(
          head_lib._summary_key(  # pylint:disable=protected-access
              self._name, metric_keys.MetricKeys.LOSS_MEAN),
          losses.compute_weighted_loss(
              unweighted_loss, weights=weights,
              reduction=losses.Reduction.MEAN))
    return model_fn.EstimatorSpec(
        mode=model_fn.ModeKeys.TRAIN,
        predictions=predictions,
        loss=training_loss,
        train_op=train_op_fn(training_loss))

  def testTrainWithTrace(self):
    logdir = os.path.join(
        tempfile.mkdtemp(prefix=self.get_temp_dir()), 'tmp_logs')
    with ops.Graph().as_default():
      random_seed.set_random_seed(0)
      tf_inputs = constant_op.constant(self._inputs, dtype=dtypes.float32)
      tf_labels = constant_op.constant(self._labels, dtype=dtypes.float32)

      tf_predictions = LogisticClassifier(tf_inputs)
      loss_ops.log_loss(tf_predictions, tf_labels)
      total_loss = loss_ops.get_total_loss()
      summary.scalar('total_loss', total_loss)

      optimizer = gradient_descent.GradientDescentOptimizer(learning_rate=1.0)

      train_op = learning.create_train_op(total_loss, optimizer)

      loss = learning.train(
          train_op,
          logdir,
          number_of_steps=300,
          log_every_n_steps=10,
          trace_every_n_steps=100)
    self.assertIsNotNone(loss)
    for trace_step in [1, 101, 201]:
      trace_filename = 'tf_trace-%d.json' % trace_step
      self.assertTrue(os.path.isfile(os.path.join(logdir, trace_filename)))

    def _tree_train_op_fn(loss):
      """Returns the op to optimize the loss."""
      if dnn_to_tree_distillation_param:
        loss_weight, loss_fn = dnn_to_tree_distillation_param
        weight_tensor = head_lib._weight_tensor(  # pylint: disable=protected-access
            features, head.weight_column_name)
        dnn_logits_fixed = array_ops.stop_gradient(dnn_logits)

        if loss_fn is None:
          # we create the loss_fn similar to the head loss_fn for
          # multi_class_head used previously as the default one.
          n_classes = 2 if head.logits_dimension == 1 else head.logits_dimension
          loss_fn = distillation_loss.create_dnn_to_tree_cross_entropy_loss_fn(
              n_classes)

        dnn_to_tree_distillation_loss = loss_weight * loss_fn(
            dnn_logits_fixed, tree_logits, weight_tensor)
        summary.scalar("dnn_to_tree_distillation_loss",
                       dnn_to_tree_distillation_loss)
        loss += dnn_to_tree_distillation_loss

      update_op = gbdt_model.train(loss, predictions_dict, labels)
      with ops.control_dependencies(
          [update_op]), (ops.colocate_with(global_step)):
        update_op = state_ops.assign_add(global_step, 1).op
        return update_op

  def create_estimator_spec(
      self, features, mode, logits, labels=None, optimizer=None,
      train_op_fn=None):
    """See `_Head`."""
    if isinstance(logits, dict):
      logits_dict = logits
    else:
      logits_dict = self._split_logits(logits)
    if labels and not isinstance(labels, dict):
      raise ValueError('labels must be a dict. Given: {}'.format(labels))

    all_estimator_spec = []
    for head in self._heads:
      head_name = head.name
      all_estimator_spec.append(
          head.create_estimator_spec(
              features=features,
              mode=mode,
              logits=logits_dict[head_name],
              labels=labels[head_name] if labels else None,
              train_op_fn=_no_op_train_fn))

    if mode == model_fn.ModeKeys.TRAIN:
      spec = self._merge_train(
          all_estimator_spec=all_estimator_spec,
          optimizer=optimizer,
          train_op_fn=train_op_fn)
      with ops.name_scope(''):
        summary.scalar(metric_keys.MetricKeys.LOSS, spec.loss)
      return spec
    if mode == model_fn.ModeKeys.PREDICT:
      return self._merge_predict(all_estimator_spec)
    if mode == model_fn.ModeKeys.EVAL:
      return self._merge_eval(all_estimator_spec)
    raise ValueError('mode={} unrecognized'.format(mode))

 def _summary_computed():
   with ops.Graph().as_default():
     sv = supervisor.Supervisor(is_chief=False)
     sess = sv.prepare_or_wait_for_session("")
     summary.scalar("c1", constant_op.constant(1))
     summary.scalar("c2", constant_op.constant(2))
     summ = summary.merge_all()
     sv.summary_computed(sess, sess.run(summ))

 def create_loss(self, features, mode, logits=None, labels=None):
   """See `_Head`."""
   model_outputs = self.state_manager.define_loss(
       self.model, features, mode)
   summary.scalar(
       head_lib._summary_key(self._name, metric_keys.MetricKeys.LOSS),
       model_outputs.loss)
   return model_outputs