我们从Python开源项目中，提取了以下9个代码示例，用于说明如何使用tensorflow.python.ops.array_ops.sequence_mask()。
def __init__(self, attention_units, memory, sequence_length=None, time_major=True, mode=0): self.attention_units = attention_units self.enc_units = memory.get_shape()[-1].value if time_major: memory = tf.transpose(memory, perm=(1, 0, 2)) self.enc_length = tf.shape(memory)[1] self.batch_size = tf.shape(memory)[0] self.mode = mode self.mask = array_ops.sequence_mask(sequence_length, self.enc_length, tf.float32) if sequence_length is not None else None self.memory = tf.reshape(memory, (tf.shape(memory)[0], self.enc_length, 1, self.enc_units)) # pre-compute Uahj to minimize the computational cost with tf.variable_scope('attention'): Ua = tf.get_variable(name='Ua', shape=(1, 1, self.enc_units, self.attention_units)) self.hidden_feats = tf.nn.conv2d(self.memory, Ua, [1, 1, 1, 1], "SAME")
def __init__(self, attention_units, memory, sequence_length=None, time_major=True, mode=0): self.attention_units = attention_units self.enc_units = memory.get_shape()[-1].value if time_major: memory = tf.transpose(memory, perm=(1,0,2)) self.enc_length = tf.shape(memory)[1] self.batch_size = tf.shape(memory)[0] self.mode = mode self.mask = array_ops.sequence_mask(sequence_length, self.enc_length) if sequence_length is not None else None self.tiny = -math.inf * tf.ones(shape=(self.batch_size, self.enc_length)) self.memory = tf.reshape(memory, (tf.shape(memory)[0], self.enc_length, 1, self.enc_units)) ### pre-compute Uahj to minimize the computational cost with tf.variable_scope('attention'): Ua = tf.get_variable(name='Ua', shape=(1, 1, self.enc_units, self.attention_units)) self.hidden_feats = tf.nn.conv2d(self.memory, Ua, [1,1,1,1], "SAME")
def __init__(self, attention_units, memory, sequence_length=None, time_major=True): self.attention_units = attention_units self.enc_units = memory.get_shape()[-1].value if time_major: memory = tf.transpose(memory, perm=(1, 0, 2)) self.enc_length = tf.shape(memory)[1] self.batch_size = tf.shape(memory)[0] self.mask = array_ops.sequence_mask(sequence_length, self.enc_length, tf.float32) if sequence_length is not None else None self.memory = tf.reshape(memory, (tf.shape(memory)[0], self.enc_length, self.enc_units))
def _maybe_mask_score(score, memory_sequence_length, score_mask_value): if memory_sequence_length is None: return score message = ("All values in memory_sequence_length must greater than zero.") with ops.control_dependencies( [check_ops.assert_positive(memory_sequence_length, message=message)]): score_mask = array_ops.sequence_mask( memory_sequence_length, maxlen=array_ops.shape(score)[1]) score_mask_values = score_mask_value * array_ops.ones_like(score) return array_ops.where(score_mask, score, score_mask_values)
def __call__(self, inputs, sequence_length=None, is_training=True, time_major=None): assert time_major is not None, "[*] You must specify whether is time_major or not!" if time_major: inputs = tf.transpose(inputs, perm=(1,0,2)) # Use batch major data. assert inputs.get_shape()[-1] == self.proj_unit[1], "[!] input's shape is not the same as ConvProj's output!" ### for correctness. if sequence_length is not None: mask = tf.expand_dims(array_ops.sequence_mask(sequence_length, tf.shape(inputs)[1], tf.float32), -1) inputs = inputs * mask ConvBankWithPool = Conv1dBankWithMaxPool(self.bank_K) ConvProj = Conv1dProjection(self.proj_unit) Highway = FCHighwayNet(self.highway_layers) rnn_cell_fw = GRUCell(self.proj_unit[1]) rnn_cell_bw = GRUCell(self.proj_unit[1]) ### calculate # conv net output_0 = ConvBankWithPool(inputs, is_training) ### for correctness. if sequence_length is not None: output_0 = output_0 * mask output_1 = ConvProj(output_0, is_training) # residual connect res_output = tf.identity(inputs) + output_1 # highway net highway_output = Highway(res_output) # biGRU # batch major final_output, *_ = bidirectional_dynamic_rnn(rnn_cell_fw, rnn_cell_bw, highway_output, sequence_length=sequence_length, time_major=False, dtype=tf.float32) final_output = tf.concat(final_output, axis=-1) if time_major: final_output = tf.transpose(final_output, perm=(1,0,2)) return final_output
def mask_activations_and_labels(activations, labels, sequence_lengths): """Remove entries outside `sequence_lengths` and returned flattened results. Args: activations: Output of the RNN, shape `[batch_size, padded_length, k]`. labels: Label values, shape `[batch_size, padded_length]`. sequence_lengths: A `Tensor` of shape `[batch_size]` with the unpadded length of each sequence. If `None`, then each sequence is unpadded. Returns: activations_masked: `logit` values with those beyond `sequence_lengths` removed for each batch. Batches are then concatenated. Shape `[tf.sum(sequence_lengths), k]` if `sequence_lengths` is not `None` and shape `[batch_size * padded_length, k]` otherwise. labels_masked: Label values after removing unneeded entries. Shape `[tf.sum(sequence_lengths)]` if `sequence_lengths` is not `None` and shape `[batch_size * padded_length]` otherwise. """ with ops.name_scope('mask_activations_and_labels', values=[activations, labels, sequence_lengths]): labels_shape = array_ops.shape(labels) batch_size = labels_shape[0] padded_length = labels_shape[1] if sequence_lengths is None: flattened_dimension = padded_length * batch_size activations_masked = array_ops.reshape(activations, [flattened_dimension, -1]) labels_masked = array_ops.reshape(labels, [flattened_dimension]) else: mask = array_ops.sequence_mask(sequence_lengths, padded_length) activations_masked = array_ops.boolean_mask(activations, mask) labels_masked = array_ops.boolean_mask(labels, mask) return activations_masked, labels_masked
def __call__(self, inputs, sequence_length=None, is_training=True, time_major=None): assert time_major is not None, "[*] You must specify whether is time_major or not!" if time_major: inputs = tf.transpose(inputs, perm=(1, 0, 2)) # Use batch major data. assert inputs.get_shape()[-1] == self.proj_unit[1], "[!] input's shape is not the same as ConvProj's output!" ### for correctness. if sequence_length is not None: mask = tf.expand_dims(array_ops.sequence_mask(sequence_length, tf.shape(inputs)[1], tf.float32), -1) inputs = inputs * mask ConvBankWithPool = Conv1dBankWithMaxPool(self.bank_K) ConvProj = Conv1dProjection(self.proj_unit) Highway = FCHighwayNet(self.highway_layers) cell = GRUCell(self.proj_unit[1]) fw_cell = FusedRNNCellAdaptor(cell) bw_cell = TimeReversedFusedRNN(fw_cell) ### calculate # conv net output_0 = ConvBankWithPool(inputs, is_training) ### for correctness. if sequence_length is not None: output_0 = output_0 * mask output_1 = ConvProj(output_0, is_training) # residual connect res_output = tf.identity(inputs) + output_1 # highway net highway_output = Highway(res_output) # biGRU # time major bGRUinp = tf.transpose(highway_output, perm=(1, 0, 2)) fw_out, _ = fw_cell(bGRUinp, sequence_length=sequence_length, scope="fw", dtype=tf.float32) bw_out, _ = bw_cell(bGRUinp, sequence_length=sequence_length, scope="bw", dtype=tf.float32) final_output = tf.concat([fw_out, bw_out], axis=-1) if not time_major: final_output = tf.transpose(final_output, perm=(1,0,2)) return final_output
