Python v1.keras方法代码示例

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import keras [as 别名]
def ctc_symbol_loss(top_out, targets, model_hparams, vocab_size, weight_fn):
  """Compute the CTC loss."""
  del model_hparams, vocab_size  # unused arg
  logits = top_out
  with tf.name_scope("ctc_loss", values=[logits, targets]):
    # For CTC we assume targets are 1d, [batch, length, 1, 1] here.
    targets_shape = targets.get_shape().as_list()
    assert len(targets_shape) == 4
    assert targets_shape[2] == 1
    assert targets_shape[3] == 1
    targets = tf.squeeze(targets, axis=[2, 3])
    logits = tf.squeeze(logits, axis=[2, 3])
    targets_mask = 1 - tf.to_int32(tf.equal(targets, 0))
    targets_lengths = tf.reduce_sum(targets_mask, axis=1)
    sparse_targets = tf.keras.backend.ctc_label_dense_to_sparse(
        targets, targets_lengths)
    xent = tf.nn.ctc_loss(
        sparse_targets,
        logits,
        targets_lengths,
        time_major=False,
        preprocess_collapse_repeated=False,
        ctc_merge_repeated=False)
    weights = weight_fn(targets)
    return tf.reduce_sum(xent), tf.reduce_sum(weights) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import keras [as 别名]
def call(self, state):
    """Creates the output tensor/op given the input state tensor.

    See https://www.tensorflow.org/api_docs/python/tf/keras/Model for more
    information on this. Note that tf.keras.Model implements `call` which is
    wrapped by `__call__` function by tf.keras.Model.

    Args:
      state: Tensor, input tensor.

    Returns:
      collections.namedtuple, output ops (graph mode) or output tensors (eager).
    """
    net = tf.cast(state, tf.float32)
    net = tf.div(net, 255.)
    net = self.conv1(net)
    net = self.conv2(net)
    net = self.conv3(net)
    net = self.flatten(net)
    net = self.dense1(net)
    net = self.dense2(net)
    unordered_q_heads = tf.reshape(net, [-1, self.num_actions, self.num_heads])
    q_heads, q_values = combine_q_functions(
        unordered_q_heads, self._transform_strategy, **self._kwargs)
    return MultiHeadNetworkType(q_heads, unordered_q_heads, q_values) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import keras [as 别名]
def video_raw_top(body_output, targets, model_hparams, vocab_size):
  del targets, model_hparams, vocab_size  # unused arg
  frames = body_output
  if isinstance(body_output, list):
    frames = tf.stack(body_output, axis=1)
  rgb_frames = common_layers.convert_real_to_rgb(frames)
  common_video.gif_summary("body_output", rgb_frames)
  return tf.expand_dims(rgb_frames, axis=-1)


# Utility functions similar to tf.keras for default transformations 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import keras [as 别名]
def __init__(self, dims):
    super(Decoder, self).__init__()
    self._decoder = keras.Sequential()
    layer_sizes = [1024, 1024, 2048]
    for layer_size in layer_sizes:
      self._decoder.add(
          keras.layers.Dense(layer_size, activation=tf.nn.leaky_relu))
    self._decoder.add(keras.layers.Dense(dims, activation=None)) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import keras [as 别名]
def __init__(self, feature_dims):
    super(Resnet18, self).__init__()
    self.conv1 = keras.layers.Conv2D(
        64, 7, strides=2, padding='same', use_bias=False)
    self.bn1 = keras.layers.BatchNormalization()
    self.relu1 = keras.layers.ReLU()
    self.maxpool = keras.layers.MaxPooling2D(3, strides=2, padding='same')
    layers = [2, 2, 2, 2]

    self.layer1 = ResLayer(BasicBlock, 64, 64, layers[0])
    self.layer2 = ResLayer(BasicBlock, 64, 128, layers[1], stride=2)
    self.layer3 = ResLayer(BasicBlock, 128, 256, layers[2], stride=2)
    self.layer4 = ResLayer(BasicBlock, 256, 512, layers[3], stride=2)

    self.fc = keras.layers.Dense(feature_dims, activation=None) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import keras [as 别名]
def __init__(self, num_actions: int, name: str = None):
    """Creates the layers used for calculating Q-values.

    Args:
      num_actions: number of actions.
      name: used to create scope for network parameters.
    """
    super(NatureDQNNetwork, self).__init__(name=name)

    self.num_actions = num_actions
    # Defining layers.
    activation_fn = tf.keras.activations.relu
    # Setting names of the layers manually to make variable names more similar
    # with tf.slim variable names/checkpoints.
    self.conv1 = tf.keras.layers.Conv2D(
        32, [8, 8],
        strides=4,
        padding='same',
        activation=activation_fn,
        name='Conv')
    self.conv2 = tf.keras.layers.Conv2D(
        64, [4, 4],
        strides=2,
        padding='same',
        activation=activation_fn,
        name='Conv')
    self.conv3 = tf.keras.layers.Conv2D(
        64, [3, 3],
        strides=1,
        padding='same',
        activation=activation_fn,
        name='Conv')
    self.flatten = tf.keras.layers.Flatten()
    self.dense1 = tf.keras.layers.Dense(
        512, activation=activation_fn, name='fully_connected')
    self.dense2 = tf.keras.layers.Dense(num_actions, name='fully_connected') 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import keras [as 别名]
def inference_network_fn(self,
                           features,
                           labels,
                           mode,
                           config=None,
                           params=None):
    """See base class documentation."""
    del mode, config, params
    if self._multi_dataset:
      net = features.x1 + features.x2
    else:
      net = features.x
    for pos, activations in enumerate([32, 16, 8]):
      # tf.keras does not support variable_scope and custom_getter.
      # Therefore, we cannot use this api yet for meta learning models.

      # Note, we have to add the MockTFModel name in order to support legacy
      # model loading.
      net = tf.layers.dense(
          net,
          units=activations,
          activation=tf.nn.elu,
          name='MockT2RModel.dense.{}'.format(pos))
      net = tf.layers.batch_normalization(
          net, name='MockT2RModel.batch_norm.{}'.format(pos))
    net = tf.layers.dense(net, units=1, name='MockT2RModel.dense.4')
    inference_outputs = {}
    inference_outputs['logit'] = net
    return inference_outputs 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import keras [as 别名]
def model_train_fn(self,
                     features,
                     labels,
                     inference_outputs,
                     mode,
                     config=None,
                     params=None):
    """See base class documentation."""
    loss = tf.keras.losses.categorical_hinge(
        y_true=labels.y, y_pred=inference_outputs['logit'])
    return tf.reduce_mean(loss) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import keras [as 别名]
def _build_model(self):
    self._model = tf.keras.Sequential()
    if not self._multi_dataset:
      for pos, activations in enumerate([32, 16, 8]):
        self._model.add(
            tf.keras.layers.Dense(
                units=activations,
                activation=tf.keras.activations.elu,
                name='MockTF2T2RModel.dense.{}'.format(pos)))
        self._model.add(
            tf.keras.layers.BatchNormalization(
                name='MockTF2T2RModel.batch_norm.{}'.format(pos)))
    self._model.add(
        tf.keras.layers.Dense(units=1, name='MockTF2T2RModel.dense.4')) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import keras [as 别名]
def main():
    parser = argparse.ArgumentParser(description='tf.keras model FLOPs & PARAMs checking tool')
    parser.add_argument('--model_path', help='model file to evaluate', type=str, required=True)
    args = parser.parse_args()

    custom_object_dict = get_custom_objects()
    model = load_model(args.model_path, compile=False, custom_objects=custom_object_dict)

    get_flops(model)