Python backend.int_shape函数代码示例

  def compute_output_shape(self, input_shape):
    input_shape = tuple(tensor_shape.TensorShape(input_shape).as_list())

    if self._output_shape is None:
      if context.executing_eagerly():
        raise NotImplementedError
      x = K.placeholder(shape=input_shape)
      x = self.call(x)
      if isinstance(x, list):
        return [tensor_shape.TensorShape(K.int_shape(x_elem)) for x_elem in x]
      else:
        return tensor_shape.TensorShape(K.int_shape(x))
    elif isinstance(self._output_shape, (tuple, list)):
      if isinstance(input_shape, list):
        num_samples = input_shape[0][0]
      else:
        num_samples = input_shape[0] if input_shape else None
      return tensor_shape.TensorShape((num_samples,) +
                                      tuple(self._output_shape))
    else:
      shape = self._output_shape(input_shape)
      if not isinstance(shape, (list, tuple)):
        raise ValueError(
            '`output_shape` function must return a tuple or a list of tuples.')
      if isinstance(shape, list):
        if isinstance(shape[0], int) or shape[0] is None:
          shape = tuple(shape)
      return tensor_shape.TensorShape(shape)

  def set_model(self, model):
    """Sets Keras model and creates summary ops."""

    self.model = model
    self.sess = K.get_session()
    # only make histogram summary op if it hasn't already been made
    if self.histogram_freq and self.merged is None:
      for layer in self.model.layers:
        for weight in layer.weights:
          mapped_weight_name = weight.name.replace(':', '_')
          tf_summary.histogram(mapped_weight_name, weight)
          if self.write_images:
            w_img = array_ops.squeeze(weight)
            shape = K.int_shape(w_img)
            if len(shape) == 2:  # dense layer kernel case
              if shape[0] > shape[1]:
                w_img = array_ops.transpose(w_img)
                shape = K.int_shape(w_img)
              w_img = array_ops.reshape(w_img, [1, shape[0], shape[1], 1])
            elif len(shape) == 3:  # convnet case
              if K.image_data_format() == 'channels_last':
                # switch to channels_first to display
                # every kernel as a separate image
                w_img = array_ops.transpose(w_img, perm=[2, 0, 1])
                shape = K.int_shape(w_img)
              w_img = array_ops.reshape(w_img,
                                        [shape[0], shape[1], shape[2], 1])
            elif len(shape) == 1:  # bias case
              w_img = array_ops.reshape(w_img, [1, shape[0], 1, 1])
            else:
              # not possible to handle 3D convnets etc.
              continue

            shape = K.int_shape(w_img)
            assert len(shape) == 4 and shape[-1] in [1, 3, 4]
            tf_summary.image(mapped_weight_name, w_img)

        if self.write_grads:
          for weight in layer.trainable_weights:
            mapped_weight_name = weight.name.replace(':', '_')
            grads = model.optimizer.get_gradients(model.total_loss, weight)

            def is_indexed_slices(grad):
              return type(grad).__name__ == 'IndexedSlices'

            grads = [grad.values if is_indexed_slices(grad) else grad
                     for grad in grads]
            tf_summary.histogram('{}_grad'.format(mapped_weight_name), grads)

        if hasattr(layer, 'output'):
          tf_summary.histogram('{}_out'.format(layer.name), layer.output)
    self.merged = tf_summary.merge_all()

    if self.write_graph:
      self.writer = self._writer_class(self.log_dir, self.sess.graph)
    else:
      self.writer = self._writer_class(self.log_dir)

def sparse_categorical_accuracy(y_true, y_pred):
  # If the shape of y_true is (num_samples, 1), squeeze to (num_samples,)
  if (len(K.int_shape(y_true)) == len(K.int_shape(y_pred))):
    y_true = array_ops.squeeze(y_true, [-1])
  y_pred = math_ops.argmax(y_pred, axis=-1)

  # If the predicted output and actual output types don't match, force cast them
  # to match.
  if K.dtype(y_pred) != K.dtype(y_true):
    y_pred = math_ops.cast(y_pred, K.dtype(y_true))

  return math_ops.cast(math_ops.equal(y_true, y_pred), K.floatx())

def sparse_categorical_accuracy(y_true, y_pred):
  # If the shape of y_true is (num_samples, 1), squeeze to (num_samples,)
  if (len(K.int_shape(y_true)) == len(K.int_shape(y_pred))):
    y_true = array_ops.squeeze(y_true, [-1])
  y_pred = math_ops.argmax(y_pred, axis=-1)

  # If the expected labels are float, we need to cast the int returned by
  # argmax to compare.
  if K.dtype(y_true) == K.floatx():
    y_pred = math_ops.cast(y_pred, K.floatx())

  return math_ops.cast(math_ops.equal(y_true, y_pred), K.floatx())

  def __call__(self, inputs, initial_state=None, constants=None, **kwargs):
    inputs, initial_state, constants = _standardize_args(
        inputs, initial_state, constants, self._num_constants)

    if initial_state is None and constants is None:
      return super(ConvRNN2D, self).__call__(inputs, **kwargs)

    # If any of `initial_state` or `constants` are specified and are Keras
    # tensors, then add them to the inputs and temporarily modify the
    # input_spec to include them.

    additional_inputs = []
    additional_specs = []
    if initial_state is not None:
      kwargs['initial_state'] = initial_state
      additional_inputs += initial_state
      self.state_spec = []
      for state in initial_state:
        shape = K.int_shape(state)
        self.state_spec.append(InputSpec(shape=shape))

      additional_specs += self.state_spec
    if constants is not None:
      kwargs['constants'] = constants
      additional_inputs += constants
      self.constants_spec = [InputSpec(shape=K.int_shape(constant))
                             for constant in constants]
      self._num_constants = len(constants)
      additional_specs += self.constants_spec
    # at this point additional_inputs cannot be empty
    for tensor in additional_inputs:
      if K.is_keras_tensor(tensor) != K.is_keras_tensor(additional_inputs[0]):
        raise ValueError('The initial state or constants of an RNN'
                         ' layer cannot be specified with a mix of'
                         ' Keras tensors and non-Keras tensors')

    if K.is_keras_tensor(additional_inputs[0]):
      # Compute the full input spec, including state and constants
      full_input = [inputs] + additional_inputs
      full_input_spec = self.input_spec + additional_specs
      # Perform the call with temporarily replaced input_spec
      original_input_spec = self.input_spec
      self.input_spec = full_input_spec
      output = super(ConvRNN2D, self).__call__(full_input, **kwargs)
      self.input_spec = original_input_spec
      return output
    else:
      return super(ConvRNN2D, self).__call__(inputs, **kwargs)

 def compute_output_shape(self, input_shape):
   if self._output_shape is None:
     if context.executing_eagerly():
       # Make use of existing autocomputation for Eager mode but provide
       # Lambda-specific error message.
       try:
         return super(Lambda, self).compute_output_shape(input_shape)
       except NotImplementedError:
         raise NotImplementedError('We could not automatically infer '
                                   'the static shape of the Lambda\'s output.'
                                   ' Please specify the `output_shape` for'
                                   ' this Lambda.')
     if isinstance(input_shape, list):
       x = [K.placeholder(shape=shape) for shape in input_shape]
     else:
       x = K.placeholder(shape=input_shape)
     x = self.call(x)
     if isinstance(x, list):
       return [tensor_shape.TensorShape(K.int_shape(x_elem)) for x_elem in x]
     else:
       return tensor_shape.TensorShape(K.int_shape(x))
   elif isinstance(self._output_shape, (tuple, list)):
     if isinstance(input_shape, list):
       num_samples = input_shape[0][0]
     else:
       num_samples = input_shape[0] if input_shape else None
     # List here represents multiple outputs.
     if isinstance(self._output_shape, list):
       return [
           tensor_shape.TensorShape((num_samples,) + tuple(single_shape))
           for single_shape in self._output_shape
       ]
     return tensor_shape.TensorShape((num_samples,) + self._output_shape)
   else:
     shape = self._output_shape(input_shape)
     if not isinstance(shape, (list, tuple)):
       raise ValueError(
           '`output_shape` function must return a tuple or a list of tuples.')
     # List here can represent multiple outputs or single output.
     if isinstance(shape, list):
       # Convert list representing single output into a tuple.
       if isinstance(shape[0], (int, type(None))):
         shape = tuple(shape)
       else:
         return [
             tensor_shape.TensorShape(single_shape) for single_shape in shape
         ]
     return tensor_shape.TensorShape(shape)

  def get_updates(self, loss, params):
    grads = self.get_gradients(loss, params)
    shapes = [K.int_shape(p) for p in params]
    accumulators = [K.zeros(shape) for shape in shapes]
    self.weights = accumulators
    self.updates = [state_ops.assign_add(self.iterations, 1)]

    lr = self.lr
    if self.initial_decay > 0:
      lr = lr * (  # pylint: disable=g-no-augmented-assignment
          1. /
          (1. +
           self.decay * math_ops.cast(self.iterations, K.dtype(self.decay))))

    for p, g, a in zip(params, grads, accumulators):
      new_a = a + math_ops.square(g)  # update accumulator
      self.updates.append(state_ops.assign(a, new_a))
      new_p = p - lr * g / (K.sqrt(new_a) + self.epsilon)

      # Apply constraints.
      if getattr(p, 'constraint', None) is not None:
        new_p = p.constraint(new_p)

      self.updates.append(state_ops.assign(p, new_p))
    return self.updates

  def _get_shape_tuple(self, init_tuple, tensor, start_idx, int_shape=None):
    """Finds non-specific dimensions in the static shapes.

    The static shapes are replaced with the corresponding dynamic shapes of the
    tensor.

    Arguments:
      init_tuple: a tuple, the first part of the output shape
      tensor: the tensor from which to get the (static and dynamic) shapes
        as the last part of the output shape
      start_idx: int, which indicate the first dimension to take from
        the static shape of the tensor
      int_shape: an alternative static shape to take as the last part
        of the output shape

    Returns:
      The new int_shape with the first part from init_tuple
      and the last part from either `int_shape` (if provided)
      or `tensor.shape`, where every `None` is replaced by
      the corresponding dimension from `tf.shape(tensor)`.
    """
    # replace all None in int_shape by K.shape
    if int_shape is None:
      int_shape = K.int_shape(tensor)[start_idx:]
    if not any(not s for s in int_shape):
      return init_tuple + tuple(int_shape)
    shape = K.shape(tensor)
    int_shape = list(int_shape)
    for i, s in enumerate(int_shape):
      if not s:
        int_shape[i] = shape[start_idx + i]
    return init_tuple + tuple(int_shape)

  def get_updates(self, loss, params):
    grads = self.get_gradients(loss, params)
    self.updates = [state_ops.assign_add(self.iterations, 1)]

    lr = self.lr
    if self.initial_decay > 0:
      lr = lr * (  # pylint: disable=g-no-augmented-assignment
          1. / (1. + self.decay * math_ops.cast(self.iterations,
                                                K.dtype(self.decay))))
    # momentum
    shapes = [K.int_shape(p) for p in params]
    moments = [K.zeros(shape) for shape in shapes]
    self.weights = [self.iterations] + moments
    for p, g, m in zip(params, grads, moments):
      v = self.momentum * m - lr * g  # velocity
      self.updates.append(state_ops.assign(m, v))

      if self.nesterov:
        new_p = p + self.momentum * v - lr * g
      else:
        new_p = p + v

      # Apply constraints.
      if getattr(p, 'constraint', None) is not None:
        new_p = p.constraint(new_p)

      self.updates.append(state_ops.assign(p, new_p))
    return self.updates

    def opt_variable(value, dtype=None, name=None, constraint=None):
      """Instantiates a variable and returns it."""
      if dtype is None:
        dtype = backend.floatx()

      variables = []
      for i in range(num_replicas):
        # Keras holds the variables in optimizer class instance , so the name
        # does not matter here. ResourceVariable constructor will find a unique
        # name (including name=None) for each replica.
        with ops.device("device:TPU:{}".format(i)):
          v = resource_variable_ops.ResourceVariable(
              value,
              dtype=dtypes_module.as_dtype(dtype),
              name=name,
              constraint=constraint)
          variables.append(v)
      name = "replicate_{}_{}".format("variable" if name is None else name,
                                      ops.uid())
      v = ReplicatedVariable(name, variables)

      # pylint: disable=protected-access

      if isinstance(value, np.ndarray):
        v._keras_shape = value.shape
      elif hasattr(value, "shape"):
        v._keras_shape = backend.int_shape(value)
      v._uses_learning_phase = False
      backend.track_variable(v)
      return v

  def call(self, inputs, training=None, mask=None):
    kwargs = {}
    if generic_utils.has_arg(self.layer.call, 'training'):
      kwargs['training'] = training
    uses_learning_phase = False  # pylint: disable=redefined-outer-name

    input_shape = K.int_shape(inputs)
    if input_shape[0]:
      # batch size matters, use rnn-based implementation
      def step(x, _):
        global uses_learning_phase  # pylint: disable=global-variable-undefined
        output = self.layer.call(x, **kwargs)
        if hasattr(output, '_uses_learning_phase'):
          uses_learning_phase = (output._uses_learning_phase or
                                 uses_learning_phase)
        return output, []

      _, outputs, _ = K.rnn(
          step,
          inputs,
          initial_states=[],
          input_length=input_shape[1],
          unroll=False)
      y = outputs
    else:
      # No batch size specified, therefore the layer will be able
      # to process batches of any size.
      # We can go with reshape-based implementation for performance.
      input_length = input_shape[1]
      if not input_length:
        input_length = array_ops.shape(inputs)[1]
      inner_input_shape = self._get_shape_tuple((-1,), inputs, 2)
      # Shape: (num_samples * timesteps, ...). And track the
      # transformation in self._input_map.
      input_uid = generic_utils.object_list_uid(inputs)
      inputs = array_ops.reshape(inputs, inner_input_shape)
      self._input_map[input_uid] = inputs
      # (num_samples * timesteps, ...)
      if generic_utils.has_arg(self.layer.call, 'mask') and mask is not None:
        inner_mask_shape = self._get_shape_tuple((-1,), mask, 2)
        kwargs['mask'] = K.reshape(mask, inner_mask_shape)
      y = self.layer.call(inputs, **kwargs)
      if hasattr(y, '_uses_learning_phase'):
        uses_learning_phase = y._uses_learning_phase
      # Shape: (num_samples, timesteps, ...)
      output_shape = self.compute_output_shape(input_shape).as_list()
      output_shape = self._get_shape_tuple(
          (-1, input_length), y, 1, output_shape[2:])
      y = array_ops.reshape(y, output_shape)

    # Apply activity regularizer if any:
    if (hasattr(self.layer, 'activity_regularizer') and
        self.layer.activity_regularizer is not None):
      regularization_loss = self.layer.activity_regularizer(y)
      self.add_loss(regularization_loss, inputs)

    if uses_learning_phase:
      y._uses_learning_phase = True
    return y

  def call(self, inputs, mask=None, training=None, initial_state=None):
    # GRU does not support constants. Ignore it during process.
    inputs, initial_state, _ = self._process_inputs(inputs, initial_state, None)

    if isinstance(mask, list):
      mask = mask[0]

    input_shape = K.int_shape(inputs)
    timesteps = input_shape[0] if self.time_major else input_shape[1]

    if not self.could_use_cudnn:
      # CuDNN does not support masking, fall back to use the normal GRU.
      kwargs = {'training': training}

      def step(cell_inputs, cell_states):
        return self.cell.call(cell_inputs, cell_states, **kwargs)

      last_output, outputs, states = K.rnn(
          step,
          inputs,
          initial_state,
          constants=None,
          go_backwards=self.go_backwards,
          mask=mask,
          unroll=self.unroll,
          input_length=timesteps,
          time_major=self.time_major,
          zero_output_for_mask=self.zero_output_for_mask)
      # This is a dummy tensor for testing purpose.
      runtime = _runtime('unknown')
    else:
      last_output, outputs, runtime, states = self._defun_gru_call(
          inputs, initial_state, training, mask)

    if self.stateful:
      updates = [state_ops.assign(self.states[0], states[0])]
      self.add_update(updates, inputs)

    if self.return_sequences:
      output = outputs
    else:
      output = last_output

    if self.return_state:
      return [output] + list(states)
    elif self._return_runtime:
      return output, runtime
    else:
      return output

def transition_block(x, reduction, name):
  """A transition block.

  Arguments:
      x: input tensor.
      reduction: float, compression rate at transition layers.
      name: string, block label.

  Returns:
      output tensor for the block.
  """
  bn_axis = 3 if K.image_data_format() == 'channels_last' else 1
  x = BatchNormalization(axis=bn_axis, epsilon=1.001e-5, name=name + '_bn')(x)
  x = Activation('relu', name=name + '_relu')(x)
  x = Conv2D(
      int(K.int_shape(x)[bn_axis] * reduction),
      1,
      use_bias=False,
      name=name + '_conv')(
          x)
  x = AveragePooling2D(2, strides=2, name=name + '_pool')(x)
  return x

def normal_lstm(inputs, init_h, init_c, kernel, recurrent_kernel, bias, units,
                activation, recurrent_activation):
  input_shape = K.int_shape(inputs)
  timesteps = input_shape[1]

  def step(cell_inputs, cell_states):
    h_tm1 = cell_states[0]  # previous memory state
    c_tm1 = cell_states[1]  # previous carry state

    # Only use the second half of the bias weights.
    _, real_bias = array_ops.split(bias, 2)

    z = K.dot(cell_inputs, kernel)
    z += K.dot(h_tm1, recurrent_kernel)
    z = K.bias_add(z, real_bias)

    z0 = z[:, :units]
    z1 = z[:, units:2 * units]
    z2 = z[:, 2 * units:3 * units]
    z3 = z[:, 3 * units:]

    i = recurrent_activation(z0)
    f = recurrent_activation(z1)
    c = f * c_tm1 + i * activation(z2)
    o = recurrent_activation(z3)

    h = o * activation(c)
    return h, [h, c]

  _, outputs, new_states = K.rnn(
      step,
      inputs, [init_h, init_c],
      constants=None,
      unroll=False,
      input_length=timesteps)
  return outputs, new_states, constant_op.constant(
      'cpu', dtype=dtypes.string, name='runtime')

def _eager_metrics_fn(model, outputs, targets):
  """Calculates the metrics for each output of the given model.

  Arguments:
      model: The model on which metrics are being calculated.
      outputs: The outputs of the given model.
      targets: The predictions or targets of the given model.

  Returns:
      Returns the metric names and metric results for each output of the model.
  """
  metric_names = []
  metric_results = []
  if not isinstance(outputs, list):
    outputs = [outputs]

  if not isinstance(targets, list):
    targets = [targets]

  for i in range(len(model.outputs)):
    output_metrics = model.nested_metrics[i]
    for nested_output_metric in output_metrics:
      metric_name, metric_fn = _get_metrics_info(
          nested_output_metric, backend.int_shape(model.outputs[i]),
          model.loss_functions[i])

      if len(model.output_names) > 1:
        metric_name = model.output_names[i] + '_' + metric_name
        if metric_name not in model.metrics_names:
          model.metrics_names.append(metric_name)

      with backend.name_scope(metric_name):
        metric_result = metric_fn(targets[i], outputs[i])
        metric_names.append(metric_name)
        metric_results.append(backend.mean(metric_result))

  return metric_results