python类shape()的实例源码

text.py 文件源码 项目:AVSR-Deep-Speech 作者: pandeydivesh15 项目源码 文件源码 阅读 29 收藏 0 点赞 0 评论 0
def sparse_tuple_from(sequences, dtype=np.int32):
    r"""Creates a sparse representention of ``sequences``.
    Args:

        * sequences: a list of lists of type dtype where each element is a sequence

    Returns a tuple with (indices, values, shape)
    """
    indices = []
    values = []

    for n, seq in enumerate(sequences):
        indices.extend(zip([n]*len(seq), range(len(seq))))
        values.extend(seq)

    indices = np.asarray(indices, dtype=np.int64)
    values = np.asarray(values, dtype=dtype)
    shape = np.asarray([len(sequences), indices.max(0)[1]+1], dtype=np.int64)

    return tf.SparseTensor(indices=indices, values=values, shape=shape)
model_utils.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 27 收藏 0 点赞 0 评论 0
def SampleRandomFrames(model_input, num_frames, num_samples):
  """Samples a random set of frames of size num_samples.

  Args:
    model_input: A tensor of size batch_size x max_frames x feature_size
    num_frames: A tensor of size batch_size x 1
    num_samples: A scalar

  Returns:
    `model_input`: A tensor of size batch_size x num_samples x feature_size
  """
  batch_size = tf.shape(model_input)[0]
  frame_index = tf.cast(
      tf.multiply(
          tf.random_uniform([batch_size, num_samples]),
          tf.tile(tf.cast(num_frames, tf.float32), [1, num_samples])), tf.int32)
  batch_index = tf.tile(
      tf.expand_dims(tf.range(batch_size), 1), [1, num_samples])
  index = tf.stack([batch_index, frame_index], 2)
  return tf.gather_nd(model_input, index)
myUtils.py 文件源码 项目:A3C 作者: go2sea 项目源码 文件源码 阅读 30 收藏 0 点赞 0 评论 0
def dense(inputs, units, bias_shape, w_i, b_i=None, activation=tf.nn.relu):
    # ??tf.layers?????flatten
    # dense1 = tf.layers.dense(tf.contrib.layers.flatten(relu5), activation=tf.nn.relu, units=50)
    if not isinstance(inputs, ops.Tensor):
        inputs = ops.convert_to_tensor(inputs, dtype='float')
        # dim_list = inputs.get_shape().as_list()
        # flatten_shape = dim_list[1] if len(dim_list) <= 2 else reduce(lambda x, y: x * y, dim_list[1:])
        # reshaped = tf.reshape(inputs, [dim_list[0], flatten_shape])
    if len(inputs.shape) > 2:
        inputs = tf.contrib.layers.flatten(inputs)
    flatten_shape = inputs.shape[1]
    weights = tf.get_variable('weights', shape=[flatten_shape, units], initializer=w_i)
    dense = tf.matmul(inputs, weights)
    if bias_shape is not None:
        assert bias_shape[0] == units
        biases = tf.get_variable('biases', shape=bias_shape, initializer=b_i)
        return activation(dense + biases) if activation is not None else dense + biases
    return activation(dense) if activation is not None else dense
keras_extensions.py 文件源码 项目:onto-lstm 作者: pdasigi 项目源码 文件源码 阅读 32 收藏 0 点赞 0 评论 0
def switch(condition, then_tensor, else_tensor):
    """
    Keras' implementation of switch for tensorflow uses tf.switch which accepts only scalar conditions.
    It should use tf.select instead.
    """
    if K.backend() == 'tensorflow':
        import tensorflow as tf
        condition_shape = condition.get_shape()
        input_shape = then_tensor.get_shape()
        if condition_shape[-1] != input_shape[-1] and condition_shape[-1] == 1:
            # This means the last dim is an embedding dim. Keras does not mask this dimension. But tf wants
            # the condition and the then and else tensors to be the same shape.
            condition = K.dot(tf.cast(condition, tf.float32), tf.ones((1, input_shape[-1])))
        return tf.select(tf.cast(condition, dtype=tf.bool), then_tensor, else_tensor)
    else:
        import theano.tensor as T
        return T.switch(condition, then_tensor, else_tensor)
beam_aligner.py 文件源码 项目:almond-nnparser 作者: Stanford-Mobisocial-IoT-Lab 项目源码 文件源码 阅读 46 收藏 0 点赞 0 评论 0
def finalize(self, outputs : BeamSearchOptimizationDecoderOutput, final_state : BeamSearchOptimizationDecoderState, sequence_lengths):
        # all output fields are [max_time, batch_size, ...]
        predicted_ids = tf.contrib.seq2seq.gather_tree(
            outputs.predicted_ids, outputs.parent_ids,
            sequence_length=sequence_lengths, name='predicted_ids')
        total_loss = tf.reduce_sum(outputs.loss, axis=0, name='violation_loss')

        predicted_time = tf.shape(predicted_ids)[0]
        last_score = predicted_time-1
        with tf.name_scope('gold_score'):
            gold_score = outputs.gold_score[last_score]
        with tf.name_scope('sequence_scores'):
            sequence_scores = outputs.scores[last_score]

        return FinalBeamSearchOptimizationDecoderOutput(beam_search_decoder_output=outputs,
                                                        predicted_ids=predicted_ids,
                                                        scores=sequence_scores,
                                                        gold_score=gold_score,
                                                        gold_beam_id=final_state.gold_beam_id,
                                                        num_available_beams=final_state.num_available_beams,
                                                        total_violation_loss=total_loss), final_state
DeepSpeech.py 文件源码 项目:AVSR-Deep-Speech 作者: pandeydivesh15 项目源码 文件源码 阅读 31 收藏 0 点赞 0 评论 0
def variable_on_worker_level(name, shape, initializer):
    r'''
    Next we concern ourselves with graph creation.
    However, before we do so we must introduce a utility function ``variable_on_worker_level()``
    used to create a variable in CPU memory.
    '''
    # Use the /cpu:0 device on worker_device for scoped operations
    if len(FLAGS.ps_hosts) == 0:
        device = worker_device
    else:
        device = tf.train.replica_device_setter(worker_device=worker_device, cluster=cluster)

    with tf.device(device):
        # Create or get apropos variable
        var = tf.get_variable(name=name, shape=shape, initializer=initializer)
    return var
text_RHL.py 文件源码 项目:AVSR-Deep-Speech 作者: pandeydivesh15 项目源码 文件源码 阅读 31 收藏 0 点赞 0 评论 0
def sparse_tuple_from(sequences, dtype=np.int32):
    r"""Creates a sparse representention of ``sequences``.
    Args:

        * sequences: a list of lists of type dtype where each element is a sequence

    Returns a tuple with (indices, values, shape)
    """
    indices = []
    values = []

    for n, seq in enumerate(sequences):
        indices.extend(zip([n]*len(seq), range(len(seq))))
        values.extend(seq)

    indices = np.asarray(indices, dtype=np.int64)
    values = np.asarray(values, dtype=dtype)
    shape = np.asarray([len(sequences), indices.max(0)[1]+1], dtype=np.int64)

    return tf.SparseTensor(indices=indices, values=values, shape=shape)
DeepSpeech_RHL.py 文件源码 项目:AVSR-Deep-Speech 作者: pandeydivesh15 项目源码 文件源码 阅读 39 收藏 0 点赞 0 评论 0
def variable_on_worker_level(name, shape, initializer):
    r'''
    Next we concern ourselves with graph creation.
    However, before we do so we must introduce a utility function ``variable_on_worker_level()``
    used to create a variable in CPU memory.
    '''
    # Use the /cpu:0 device on worker_device for scoped operations
    if len(FLAGS.ps_hosts) == 0:
        device = worker_device
    else:
        device = tf.train.replica_device_setter(worker_device=worker_device, cluster=cluster)

    with tf.device(device):
        # Create or get apropos variable
        var = tf.get_variable(name=name, shape=shape, initializer=initializer)
    return var
DeepSpeech_RHL_AVSR.py 文件源码 项目:AVSR-Deep-Speech 作者: pandeydivesh15 项目源码 文件源码 阅读 33 收藏 0 点赞 0 评论 0
def variable_on_worker_level(name, shape, initializer):
    r'''
    Next we concern ourselves with graph creation.
    However, before we do so we must introduce a utility function ``variable_on_worker_level()``
    used to create a variable in CPU memory.
    '''
    # Use the /cpu:0 device on worker_device for scoped operations
    if len(FLAGS.ps_hosts) == 0:
        device = worker_device
    else:
        device = tf.train.replica_device_setter(worker_device=worker_device, cluster=cluster)

    with tf.device(device):
        # Create or get apropos variable
        var = tf.get_variable(name=name, shape=shape, initializer=initializer)
    return var
frame_level_models.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 24 收藏 0 点赞 0 评论 0
def highway(self, input_1, input_2, size_1, size_2, l2_penalty=1e-8, layer_size=1):
        output = input_2
        for idx in range(layer_size):
            with tf.name_scope('output_lin_%d' % idx):
                W = tf.Variable(tf.truncated_normal([size_2,size_1], stddev=0.1), name="W")
                b = tf.Variable(tf.constant(0.1, shape=[size_1]), name="b")
                tf.add_to_collection(name=tf.GraphKeys.REGULARIZATION_LOSSES, value=l2_penalty*tf.nn.l2_loss(W))
                tf.add_to_collection(name=tf.GraphKeys.REGULARIZATION_LOSSES, value=l2_penalty*tf.nn.l2_loss(b))
                output = tf.nn.relu(tf.nn.xw_plus_b(output,W,b))
            with tf.name_scope('transform_lin_%d' % idx):
                W = tf.Variable(tf.truncated_normal([size_1,size_1], stddev=0.1), name="W")
                b = tf.Variable(tf.constant(0.1, shape=[size_1]), name="b")
                tf.add_to_collection(name=tf.GraphKeys.REGULARIZATION_LOSSES, value=l2_penalty*tf.nn.l2_loss(W))
                tf.add_to_collection(name=tf.GraphKeys.REGULARIZATION_LOSSES, value=l2_penalty*tf.nn.l2_loss(b))
                transform_gate = tf.sigmoid(tf.nn.xw_plus_b(input_1,W,b))
            carry_gate = tf.constant(1.0) - transform_gate
            output = transform_gate * output + carry_gate * input_1
        return output
losses.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 39 收藏 0 点赞 0 评论 0
def calculate_loss_distill_boost(self, predictions, labels_distill, labels, **unused_params):
    with tf.name_scope("loss_distill_boost"):
      print("loss_distill_boost")
      epsilon = 10e-6
      float_labels = tf.cast(labels, tf.float32)
      batch_size = tf.shape(float_labels)[0]
      float_labels_distill = tf.cast(labels_distill, tf.float32)
      error = tf.negative(float_labels * tf.log(float_labels_distill + epsilon) + (
          1 - float_labels) * tf.log(1 - float_labels_distill + epsilon))
      error = tf.reduce_sum(error,axis=1,keep_dims=True)
      alpha = error / tf.reduce_sum(error) * tf.cast(batch_size,dtype=tf.float32)
      alpha = tf.clip_by_value(alpha, 0.5, 5)
      alpha = alpha / tf.reduce_sum(alpha) * tf.cast(batch_size,dtype=tf.float32)
      cross_entropy_loss = float_labels * tf.log(predictions + epsilon) + (
          1 - float_labels) * tf.log(1 - predictions + epsilon)
      cross_entropy_loss = tf.negative(cross_entropy_loss * alpha)

      return tf.reduce_mean(tf.reduce_sum(cross_entropy_loss, 1))
losses.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 27 收藏 0 点赞 0 评论 0
def calculate_loss_distill_relabel(self, predictions, labels_distill, labels, **unused_params):
    with tf.name_scope("loss_distill_relabel"):
      print("loss_distill_relabel")
      epsilon = 10e-6
      float_labels = tf.cast(labels, tf.float32)
      sum_labels = tf.cast(tf.reduce_sum(float_labels),dtype=tf.int32)
      pos_distill, _ = tf.nn.top_k(tf.reshape(labels_distill,[-1]), k=sum_labels)
      labels_true = tf.ones(tf.shape(labels))
      labels_false = tf.zeros(tf.shape(labels))
      labels_add = tf.where(tf.greater_equal(labels_distill, pos_distill[-1]), labels_true, labels_false)
      print(labels_add.get_shape().as_list())
      float_labels = float_labels+labels_add*(1.0-float_labels)
      cross_entropy_loss = float_labels * tf.log(predictions + epsilon) + (
          1 - float_labels) * tf.log(1 - predictions + epsilon)
      cross_entropy_loss = tf.negative(cross_entropy_loss)

      return tf.reduce_mean(tf.reduce_sum(cross_entropy_loss, 1))
model_utils.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 39 收藏 0 点赞 0 评论 0
def FramePooling(frames, method, **unused_params):
  """Pools over the frames of a video.

  Args:
    frames: A tensor with shape [batch_size, num_frames, feature_size].
    method: "average", "max", "attention", or "none".
  Returns:
    A tensor with shape [batch_size, feature_size] for average, max, or
    attention pooling. A tensor with shape [batch_size*num_frames, feature_size]
    for none pooling.

  Raises:
    ValueError: if method is other than "average", "max", "attention", or
    "none".
  """
  if method == "average":
    return tf.reduce_mean(frames, 1)
  elif method == "max":
    return tf.reduce_max(frames, 1)
  elif method == "none":
    feature_size = frames.shape_as_list()[2]
    return tf.reshape(frames, [-1, feature_size])
  else:
    raise ValueError("Unrecognized pooling method: %s" % method)
losses.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 29 收藏 0 点赞 0 评论 0
def calculate_loss(self, predictions, support_predictions, labels, **unused_params):
    """ 
    support_predictions batch_size x num_models x num_classes
    predictions = tf.reduce_mean(support_predictions, axis=1)
    """
    model_count = tf.shape(support_predictions)[1]
    vocab_size = tf.shape(support_predictions)[2]

    mean_predictions = tf.reduce_mean(support_predictions, axis=1, keep_dims=True)
    support_labels = tf.tile(tf.expand_dims(tf.cast(labels, dtype=tf.float32), axis=1), multiples=[1,model_count,1])
    support_means = tf.stop_gradient(tf.tile(mean_predictions, multiples=[1,model_count,1]))

    support_predictions = tf.reshape(support_predictions, shape=[-1,model_count*vocab_size])
    support_labels = tf.reshape(support_labels, shape=[-1,model_count*vocab_size])
    support_means = tf.reshape(support_means, shape=[-1,model_count*vocab_size])

    ce_loss_fn = CrossEntropyLoss()
    # The cross entropy between predictions and ground truth
    cross_entropy_loss = ce_loss_fn.calculate_loss(support_predictions, support_labels, **unused_params)
    # The cross entropy between predictions and mean predictions
    divergence = ce_loss_fn.calculate_loss(support_predictions, support_means, **unused_params)

    loss = cross_entropy_loss * (1.0 - FLAGS.support_loss_percent) - divergence * FLAGS.support_loss_percent
    return loss
model_utils.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 51 收藏 0 点赞 0 评论 0
def FramePooling(frames, method, **unused_params):
  """Pools over the frames of a video.

  Args:
    frames: A tensor with shape [batch_size, num_frames, feature_size].
    method: "average", "max", "attention", or "none".
  Returns:
    A tensor with shape [batch_size, feature_size] for average, max, or
    attention pooling. A tensor with shape [batch_size*num_frames, feature_size]
    for none pooling.

  Raises:
    ValueError: if method is other than "average", "max", "attention", or
    "none".
  """
  if method == "average":
    return tf.reduce_mean(frames, 1)
  elif method == "max":
    return tf.reduce_max(frames, 1)
  elif method == "none":
    feature_size = frames.shape_as_list()[2]
    return tf.reshape(frames, [-1, feature_size])
  else:
    raise ValueError("Unrecognized pooling method: %s" % method)
readers.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 36 收藏 0 点赞 0 评论 0
def resize_axis(tensor, axis, new_size, fill_value=0):
  tensor = tf.convert_to_tensor(tensor)
  shape = tf.unstack(tf.shape(tensor))

  pad_shape = shape[:]
  pad_shape[axis] = tf.maximum(0, new_size - shape[axis])

  shape[axis] = tf.minimum(shape[axis], new_size)
  shape = tf.stack(shape)

  resized = tf.concat([
      tf.slice(tensor, tf.zeros_like(shape), shape),
      tf.fill(tf.stack(pad_shape), tf.cast(fill_value, tensor.dtype))
  ], axis)

  # Update shape.
  new_shape = tensor.get_shape().as_list()  # A copy is being made.
  new_shape[axis] = new_size
  resized.set_shape(new_shape)
  return resized
readers.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 49 收藏 0 点赞 0 评论 0
def prepare_reader(self, filename_queue, batch_size=1024):

    reader = tf.TFRecordReader()
    _, serialized_examples = reader.read_up_to(filename_queue, batch_size)

    # set the mapping from the fields to data types in the proto
    num_features = len(self.feature_names)
    assert num_features > 0, "self.feature_names is empty!"
    assert len(self.feature_names) == len(self.feature_sizes), \
    "length of feature_names (={}) != length of feature_sizes (={})".format( \
    len(self.feature_names), len(self.feature_sizes))

    feature_map = {"video_id": tf.FixedLenFeature([], tf.string),
                   "labels": tf.VarLenFeature(tf.int64)}
    for feature_index in range(num_features):
      feature_map[self.feature_names[feature_index]] = tf.FixedLenFeature(
          [self.feature_sizes[feature_index]], tf.float32)

    features = tf.parse_example(serialized_examples, features=feature_map)
    labels = tf.sparse_to_indicator(features["labels"], self.num_classes)
    labels.set_shape([None, self.num_classes])
    concatenated_features = tf.concat([
        features[feature_name] for feature_name in self.feature_names], 1)

    return features["video_id"], concatenated_features, labels, tf.ones([tf.shape(serialized_examples)[0]])
model_utils.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 34 收藏 0 点赞 0 评论 0
def SampleRandomFrames(model_input, num_frames, num_samples):
  """Samples a random set of frames of size num_samples.

  Args:
    model_input: A tensor of size batch_size x max_frames x feature_size
    num_frames: A tensor of size batch_size x 1
    num_samples: A scalar

  Returns:
    `model_input`: A tensor of size batch_size x num_samples x feature_size
  """
  batch_size = tf.shape(model_input)[0]
  frame_index = tf.cast(
      tf.multiply(
          tf.random_uniform([batch_size, num_samples]),
          tf.tile(tf.cast(num_frames, tf.float32), [1, num_samples])), tf.int32)
  batch_index = tf.tile(
      tf.expand_dims(tf.range(batch_size), 1), [1, num_samples])
  index = tf.stack([batch_index, frame_index], 2)
  return tf.gather_nd(model_input, index)
model_utils.py 文件源码 项目:youtube-8m 作者: wangheda 项目源码 文件源码 阅读 36 收藏 0 点赞 0 评论 0
def FramePooling(frames, method, **unused_params):
  """Pools over the frames of a video.

  Args:
    frames: A tensor with shape [batch_size, num_frames, feature_size].
    method: "average", "max", "attention", or "none".
  Returns:
    A tensor with shape [batch_size, feature_size] for average, max, or
    attention pooling. A tensor with shape [batch_size*num_frames, feature_size]
    for none pooling.

  Raises:
    ValueError: if method is other than "average", "max", "attention", or
    "none".
  """
  if method == "average":
    return tf.reduce_mean(frames, 1)
  elif method == "max":
    return tf.reduce_max(frames, 1)
  elif method == "none":
    feature_size = frames.shape_as_list()[2]
    return tf.reshape(frames, [-1, feature_size])
  else:
    raise ValueError("Unrecognized pooling method: %s" % method)
distributions.py 文件源码 项目:distributional_perspective_on_RL 作者: Kiwoo 项目源码 文件源码 阅读 40 收藏 0 点赞 0 评论 0
def sample_dtype(self):
        return tf.int32

# WRONG SECOND DERIVATIVES
# class CategoricalPd(Pd):
#     def __init__(self, logits):
#         self.logits = logits
#         self.ps = tf.nn.softmax(logits)
#     @classmethod
#     def fromflat(cls, flat):
#         return cls(flat)
#     def flatparam(self):
#         return self.logits
#     def mode(self):
#         return U.argmax(self.logits, axis=1)
#     def logp(self, x):
#         return -tf.nn.sparse_softmax_cross_entropy_with_logits(self.logits, x)
#     def kl(self, other):
#         return tf.nn.softmax_cross_entropy_with_logits(other.logits, self.ps) \
#                 - tf.nn.softmax_cross_entropy_with_logits(self.logits, self.ps)
#     def entropy(self):
#         return tf.nn.softmax_cross_entropy_with_logits(self.logits, self.ps)
#     def sample(self):
#         u = tf.random_uniform(tf.shape(self.logits))
#         return U.argmax(self.logits - tf.log(-tf.log(u)), axis=1)
seq2seq_model.py 文件源码 项目:seq2seq 作者: google 项目源码 文件源码 阅读 34 收藏 0 点赞 0 评论 0
def compute_loss(self, decoder_output, _features, labels):
    """Computes the loss for this model.

    Returns a tuple `(losses, loss)`, where `losses` are the per-batch
    losses and loss is a single scalar tensor to minimize.
    """
    #pylint: disable=R0201
    # Calculate loss per example-timestep of shape [B, T]
    losses = seq2seq_losses.cross_entropy_sequence_loss(
        logits=decoder_output.logits[:, :, :],
        targets=tf.transpose(labels["target_ids"][:, 1:], [1, 0]),
        sequence_length=labels["target_len"] - 1)

    # Calculate the average log perplexity
    loss = tf.reduce_sum(losses) / tf.to_float(
        tf.reduce_sum(labels["target_len"] - 1))

    return losses, loss
image_encoder.py 文件源码 项目:seq2seq 作者: google 项目源码 文件源码 阅读 31 收藏 0 点赞 0 评论 0
def encode(self, inputs):
    inputs = tf.image.resize_images(
        images=inputs,
        size=[self.params["resize_height"], self.params["resize_width"]],
        method=tf.image.ResizeMethod.BILINEAR)

    outputs, _ = inception_v3_base(tf.to_float(inputs))
    output_shape = outputs.get_shape()  #pylint: disable=E1101
    shape_list = output_shape.as_list()

    # Take attentin over output elemnts in width and height dimension:
    # Shape: [B, W*H, ...]
    outputs_flat = tf.reshape(outputs, [shape_list[0], -1, shape_list[-1]])

    # Final state is the pooled output
    # Shape: [B, W*H*...]
    final_state = tf.contrib.slim.avg_pool2d(
        outputs, output_shape[1:3], padding="VALID", scope="pool")
    final_state = tf.contrib.slim.flatten(outputs, scope="flatten")

    return EncoderOutput(
        outputs=outputs_flat,
        final_state=final_state,
        attention_values=outputs_flat,
        attention_values_length=tf.shape(outputs_flat)[1])
pooling_encoder.py 文件源码 项目:seq2seq 作者: google 项目源码 文件源码 阅读 32 收藏 0 点赞 0 评论 0
def position_encoding(sentence_size, embedding_size):
  """
  Position Encoding described in section 4.1 of
  End-To-End Memory Networks (https://arxiv.org/abs/1503.08895).

  Args:
    sentence_size: length of the sentence
    embedding_size: dimensionality of the embeddings

  Returns:
    A numpy array of shape [sentence_size, embedding_size] containing
    the fixed position encodings for each sentence position.
  """
  encoding = np.ones((sentence_size, embedding_size), dtype=np.float32)
  ls = sentence_size + 1
  le = embedding_size + 1
  for k in range(1, le):
    for j in range(1, ls):
      encoding[j-1, k-1] = (1.0 - j/float(ls)) - (
          k / float(le)) * (1. - 2. * j/float(ls))
  return encoding
hmc_samplers.py 文件源码 项目:bnn-analysis 作者: myshkov 项目源码 文件源码 阅读 33 收藏 0 点赞 0 评论 0
def _add_mh_correction(self, initial_position, initial_velocity, final_position, final_velocity):
        """ Applies MH accept/reject correction. """
        initial_energy = self._hamiltonian(initial_position, initial_velocity)
        final_energy = self._hamiltonian(final_position, final_velocity)
        accepted = self._metropolis_hastings_accept(initial_energy, final_energy)
        accepted = tf.to_float(accepted)

        # add acceptance to fetched values
        self._accepted = accepted

        if self.seek_step_sizes or self.fade_in_velocities:
            burned_in = tf.to_float(self._burn_in_ratio == 1)
            accepted = accepted * burned_in + tf.ones(shape=tf.shape(accepted)) * (1 - burned_in)

        # apply MH decision
        final_position = self._transpose_mul(final_position, accepted) + \
                         self._transpose_mul(initial_position, tf.ones(shape=tf.shape(accepted)) - accepted)

        final_velocity = self._transpose_mul(final_velocity, accepted) + \
                         self._transpose_mul(-initial_velocity, tf.ones(shape=tf.shape(accepted)) - accepted)

        return final_position, final_velocity
hmc_samplers.py 文件源码 项目:bnn-analysis 作者: myshkov 项目源码 文件源码 阅读 33 收藏 0 点赞 0 评论 0
def _leapfrog_step(self, position, velocity, velocity_step_multiplier=1.):
        """ Makes a single leapfrog step with friction. """
        d_energy = self._d_energy_fn(position)

        friction = self.friction
        deceleration = -friction * self._transpose_mul(velocity, self._current_step_size)

        velocity -= self._transpose_mul(d_energy, velocity_step_multiplier * self._current_step_size)
        velocity += deceleration

        # B_hat = 0, C = friction
        noise = tf.random_normal(tf.shape(velocity))
        stddevs = (2 * friction * self._current_step_size) ** 0.5
        noise = self._transpose_mul(noise, stddevs)

        velocity += noise

        position = position + self._transpose_mul(velocity, self._current_step_size)

        return position, velocity
word_rnn.py 文件源码 项目:deep-learning 作者: ljanyst 项目源码 文件源码 阅读 34 收藏 0 点赞 0 评论 0
def get_optimizer(self, learning_rate = 0.001):
        with tf.name_scope('loss'):
            input_shape = tf.shape(self.inputs)
            ones        = tf.ones([input_shape[0], input_shape[1]])
            loss = tf.contrib.seq2seq.sequence_loss(self.logits, self.targets,
                                                    ones)

        #-----------------------------------------------------------------------
        # Build the optimizer
        #-----------------------------------------------------------------------
        with tf.name_scope('optimizer'):
            optimizer = tf.train.AdamOptimizer(learning_rate)
            gradients = optimizer.compute_gradients(loss)
            capped_gradients = [(tf.clip_by_value(grad, -1., 1.), var) \
                                for grad, var in gradients if grad is not None]
            optimizer_op = optimizer.apply_gradients(capped_gradients)

        return optimizer_op, loss
preprocessing.py 文件源码 项目:benchmarks 作者: tensorflow 项目源码 文件源码 阅读 30 收藏 0 点赞 0 评论 0
def decode_jpeg(image_buffer, scope=None):  # , dtype=tf.float32):
  """Decode a JPEG string into one 3-D float image Tensor.

  Args:
    image_buffer: scalar string Tensor.
    scope: Optional scope for op_scope.
  Returns:
    3-D float Tensor with values ranging from [0, 1).
  """
  # with tf.op_scope([image_buffer], scope, 'decode_jpeg'):
  # with tf.name_scope(scope, 'decode_jpeg', [image_buffer]):
  with tf.name_scope(scope or 'decode_jpeg'):
    # Decode the string as an RGB JPEG.
    # Note that the resulting image contains an unknown height and width
    # that is set dynamically by decode_jpeg. In other words, the height
    # and width of image is unknown at compile-time.
    image = tf.image.decode_jpeg(image_buffer, channels=3,
                                 fancy_upscaling=False,
                                 dct_method='INTEGER_FAST')

    # image = tf.Print(image, [tf.shape(image)], 'Image shape: ')

    return image
resnet_v1.py 文件源码 项目:HandDetection 作者: YunqiuXu 项目源码 文件源码 阅读 32 收藏 0 点赞 0 评论 0
def _crop_pool_layer(self, bottom, rois, name):
    with tf.variable_scope(name) as scope:
      batch_ids = tf.squeeze(tf.slice(rois, [0, 0], [-1, 1], name="batch_id"), [1])
      # Get the normalized coordinates of bboxes
      bottom_shape = tf.shape(bottom)
      height = (tf.to_float(bottom_shape[1]) - 1.) * np.float32(self._feat_stride[0])
      width = (tf.to_float(bottom_shape[2]) - 1.) * np.float32(self._feat_stride[0])
      x1 = tf.slice(rois, [0, 1], [-1, 1], name="x1") / width
      y1 = tf.slice(rois, [0, 2], [-1, 1], name="y1") / height
      x2 = tf.slice(rois, [0, 3], [-1, 1], name="x2") / width
      y2 = tf.slice(rois, [0, 4], [-1, 1], name="y2") / height
      # Won't be back-propagated to rois anyway, but to save time
      bboxes = tf.stop_gradient(tf.concat([y1, x1, y2, x2], 1))
      if cfg.RESNET.MAX_POOL:
        pre_pool_size = cfg.POOLING_SIZE * 2
        crops = tf.image.crop_and_resize(bottom, bboxes, tf.to_int32(batch_ids), [pre_pool_size, pre_pool_size],
                                         name="crops")
        crops = slim.max_pool2d(crops, [2, 2], padding='SAME')
      else:
        crops = tf.image.crop_and_resize(bottom, bboxes, tf.to_int32(batch_ids), [cfg.POOLING_SIZE, cfg.POOLING_SIZE],
                                         name="crops")
    return crops

  # Do the first few layers manually, because 'SAME' padding can behave inconsistently
  # for images of different sizes: sometimes 0, sometimes 1
network_vgg16.py 文件源码 项目:HandDetection 作者: YunqiuXu 项目源码 文件源码 阅读 33 收藏 0 点赞 0 评论 0
def _crop_pool_layer(self, bottom, rois, name):
    with tf.variable_scope(name) as scope:
      batch_ids = tf.squeeze(tf.slice(rois, [0, 0], [-1, 1], name="batch_id"), [1])
      # Get the normalized coordinates of bounding boxes
      bottom_shape = tf.shape(bottom)
      height = (tf.to_float(bottom_shape[1]) - 1.) * np.float32(self._feat_stride[0])
      width = (tf.to_float(bottom_shape[2]) - 1.) * np.float32(self._feat_stride[0])
      x1 = tf.slice(rois, [0, 1], [-1, 1], name="x1") / width
      y1 = tf.slice(rois, [0, 2], [-1, 1], name="y1") / height
      x2 = tf.slice(rois, [0, 3], [-1, 1], name="x2") / width
      y2 = tf.slice(rois, [0, 4], [-1, 1], name="y2") / height
      # Won't be back-propagated to rois anyway, but to save time
      bboxes = tf.stop_gradient(tf.concat([y1, x1, y2, x2], axis=1))
      pre_pool_size = cfg.POOLING_SIZE * 2
      crops = tf.image.crop_and_resize(bottom, bboxes, tf.to_int32(batch_ids), [pre_pool_size, pre_pool_size], name="crops")

    return slim.max_pool2d(crops, [2, 2], padding='SAME')
network_vgg16.py 文件源码 项目:HandDetection 作者: YunqiuXu 项目源码 文件源码 阅读 34 收藏 0 点赞 0 评论 0
def _anchor_component(self):
    with tf.variable_scope('ANCHOR_' + self._tag) as scope:
      # just to get the shape right
      height = tf.to_int32(tf.ceil(self._im_info[0] / np.float32(self._feat_stride[0])))
      width = tf.to_int32(tf.ceil(self._im_info[1] / np.float32(self._feat_stride[0])))
      anchors, anchor_length = tf.py_func(generate_anchors_pre,
                                          [height, width,
                                           self._feat_stride, self._anchor_scales, self._anchor_ratios],
                                          [tf.float32, tf.int32], name="generate_anchors")
      anchors.set_shape([None, 4])
      anchor_length.set_shape([])
      self._anchors = anchors
      self._anchor_length = anchor_length

  # [Hand Detection] Batch normalization
  # http://stackoverflow.com/a/34634291/2267819
  # Note that this is different from the paper(they use another method)


问题


面经


文章

微信
公众号

扫码关注公众号