python类float32()的实例源码

def calculate_loss(self, predictions, labels, weights=None, **unused_params):
    with tf.name_scope("loss_xent"):
      epsilon = 10e-6
      if FLAGS.label_smoothing:
        float_labels = smoothing(labels)
      else:
        float_labels = tf.cast(labels, 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)
      if weights is not None:
        print cross_entropy_loss, weights
        weighted_loss = tf.einsum("ij,i->ij", cross_entropy_loss, weights)
        print "create weighted_loss", weighted_loss
        return tf.reduce_mean(tf.reduce_sum(weighted_loss, 1))
      else:
        return tf.reduce_mean(tf.reduce_sum(cross_entropy_loss, 1))

def get_video_weights(video_id_batch):
  video_id_to_index = tf.contrib.lookup.string_to_index_table_from_file(
                          vocabulary_file=FLAGS.sample_vocab_file, default_value=0)
  indexes = video_id_to_index.lookup(video_id_batch)
  weights, length = get_video_weights_array()
  weights_input = tf.placeholder(tf.float32, shape=[length], name="sample_weights_input")
  weights_tensor = tf.get_variable("sample_weights",
                               shape=[length],
                               trainable=False,
                               dtype=tf.float32,
                               initializer=tf.constant_initializer(weights))
  weights_assignment = tf.assign(weights_tensor, weights_input)

  tf.add_to_collection("weights_input", weights_input)
  tf.add_to_collection("weights_assignment", weights_assignment)

  video_weight_batch = tf.nn.embedding_lookup(weights_tensor, indexes)
  return video_weight_batch

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]])

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)

def preprocess_for_eval(image, height, width,
                        central_fraction=0.875, scope=None):
  """Prepare one image for evaluation.
  If height and width are specified it would output an image with that size by
  applying resize_bilinear.
  If central_fraction is specified it would cropt the central fraction of the
  input image.
  Args:
    image: 3-D Tensor of image. If dtype is tf.float32 then the range should be
      [0, 1], otherwise it would converted to tf.float32 assuming that the range
      is [0, MAX], where MAX is largest positive representable number for
      int(8/16/32) data type (see `tf.image.convert_image_dtype` for details)
    height: integer
    width: integer
    central_fraction: Optional Float, fraction of the image to crop.
    scope: Optional scope for name_scope.
  Returns:
    3-D float Tensor of prepared image.
  """
  with tf.name_scope(scope, 'eval_image', [image, height, width]):
    if image.dtype != tf.float32:
      image = tf.image.convert_image_dtype(image, dtype=tf.float32)
    # Crop the central region of the image with an area containing 87.5% of
    # the original image.
    if central_fraction:
      image = tf.image.central_crop(image, central_fraction=central_fraction)

    if height and width:
      # Resize the image to the specified height and width.
      image = tf.expand_dims(image, 0)
      image = tf.image.resize_bilinear(image, [height, width],
                                       align_corners=False)
      image = tf.squeeze(image, [0])
    image = tf.subtract(image, 0.5)
    image = tf.multiply(image, 2.0)
    return image

def actor_loss(self):
        if self.config.mode == 'discrete':
            log_prob = tf.reduce_sum(tf.log(self.a_prob) * tf.one_hot(self.action_input, self.action_dim, dtype=tf.float32),
                                     axis=1, keep_dims=True)
            # use entropy to encourage exploration
            exp_v = log_prob * self.TD_loss
            entropy = -tf.reduce_sum(self.a_prob * tf.log(self.a_prob), axis=1, keep_dims=True)  # encourage exploration
            exp_v = self.config.ENTROPY_BETA * entropy + exp_v
            return tf.reduce_mean(-exp_v)  # ????????log_prb????????????????????TD_loss
        elif self.config.mode == 'continuous':
            log_prob = self.action_normal_dist.log_prob(self.action_input)
            exp_v = log_prob * self.TD_loss
            # use entropy to encourage exploration
            exp_v = self.config.ENTROPY_BETA * self.action_normal_dist.entropy() + exp_v
            return tf.reduce_mean(-exp_v)

def noisy_dense(inputs, units, bias_shape, c_names, w_i, b_i=None, activation=tf.nn.relu, noisy_distribution='factorised'):
    def f(e_list):
        return tf.multiply(tf.sign(e_list), tf.pow(tf.abs(e_list), 0.5))
    # ??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)
    w_noise = tf.get_variable('w_noise', [flatten_shape, units], initializer=w_i, collections=c_names)
    if noisy_distribution == 'independent':
        weights += tf.multiply(tf.random_normal(shape=w_noise.shape), w_noise)
    elif noisy_distribution == 'factorised':
        noise_1 = f(tf.random_normal(tf.TensorShape([flatten_shape, 1]), dtype=tf.float32))  # ???????????????
        noise_2 = f(tf.random_normal(tf.TensorShape([1, units]), dtype=tf.float32))
        weights += tf.multiply(noise_1 * noise_2, w_noise)
    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)
        b_noise = tf.get_variable('b_noise', [1, units], initializer=b_i, collections=c_names)
        if noisy_distribution == 'independent':
            biases += tf.multiply(tf.random_normal(shape=b_noise.shape), b_noise)
        elif noisy_distribution == 'factorised':
            biases += tf.multiply(noise_2, b_noise)
        return activation(dense + biases) if activation is not None else dense + biases
    return activation(dense) if activation is not None else dense


# ???bias??????relu

def make_skipgram_softmax_loss(embeddings_matrix, vocabulary_size, vector_size):
    vectors = tf.get_variable('vectors', (vocabulary_size, vector_size), dtype=tf.float32, initializer=tf.constant_initializer(embeddings_matrix))
    minibatch = tf.placeholder(shape=(None, 2), dtype=tf.int32)

    center_word_vector = tf.nn.embedding_lookup(vectors, minibatch[:,0])
    yhat = tf.matmul(center_word_vector, vectors, transpose_b=True)

    predict_word = minibatch[:,1]
    loss = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=predict_word, logits=yhat)
    loss = tf.reduce_mean(loss)
    return vectors, minibatch, loss

def encode(self, inputs, input_length, _parses):
        with tf.name_scope('LSTMEncoder'):
            cell_enc = tf.contrib.rnn.MultiRNNCell([self._make_rnn_cell(i) for i in range(self._num_layers)])

            return tf.nn.dynamic_rnn(cell_enc, inputs, sequence_length=input_length,
                                     dtype=tf.float32)

def encode(self, inputs, input_length, _parses):
        with tf.name_scope('BiLSTMEncoder'):
            fw_cell_enc = tf.contrib.rnn.MultiRNNCell([self._make_rnn_cell(i) for i in range(self._num_layers)])
            bw_cell_enc = tf.contrib.rnn.MultiRNNCell([self._make_rnn_cell(i) for i in range(self._num_layers)])

            outputs, output_state = tf.nn.bidirectional_dynamic_rnn(fw_cell_enc, bw_cell_enc, inputs, input_length,
                                                                    dtype=tf.float32)

            fw_output_state, bw_output_state = output_state
            # concat each element of the final state, so that we're compatible with a unidirectional
            # decoder
            output_state = nest.pack_sequence_as(fw_output_state, [tf.concat((x, y), axis=1) for x, y in zip(nest.flatten(fw_output_state), nest.flatten(bw_output_state))])

            return tf.concat(outputs, axis=2), output_state

def encode(self, inputs, _input_length, _parses):
        with tf.variable_scope('BagOfWordsEncoder'):
            W = tf.get_variable('W', (self.embed_size, self.output_size))
            b = tf.get_variable('b', shape=(self.output_size,), initializer=tf.constant_initializer(0, tf.float32))

            enc_hidden_states = tf.tanh(tf.tensordot(inputs, W, [[2], [0]]) + b)
            enc_final_state = tf.reduce_sum(enc_hidden_states, axis=1)

            #assert enc_hidden_states.get_shape()[1:] == (self.config.max_length, self.config.hidden_size)
            if self._cell_type == 'lstm':
                enc_final_state = (tf.contrib.rnn.LSTMStateTuple(enc_final_state, enc_final_state),)

            enc_output = tf.nn.dropout(enc_hidden_states, keep_prob=self._dropout, seed=12345)

            return enc_output, enc_final_state

def initialize(self):
        """Initialize the decoder.
        Args:
          name: Name scope for any created operations.
        Returns:
          `(finished, start_inputs, initial_state)`.
        """
        start_inputs = self._embedding_fn(self._tiled_start_tokens)
        print('start_inputs', start_inputs)
        finished = tf.zeros((self.batch_size, self._beam_width), dtype=tf.bool)

        self._initial_num_available_beams = tf.ones((self._batch_size,), dtype=tf.int32)
        self._full_num_available_beams = tf.fill((self._batch_size,), self._beam_width)

        with tf.name_scope('first_beam_mask'):
            self._first_beam_mask = self._make_beam_mask(self._initial_num_available_beams)
        with tf.name_scope('full_beam_mask'):
            self._full_beam_mask = self._make_beam_mask(self._full_num_available_beams)
        with tf.name_scope('minus_inifinity_scores'):
            self._minus_inifinity_scores = tf.fill((self.batch_size, self._beam_width, self._output_size), -1e+8)

        self._batch_size_range = tf.range(self.batch_size)
        initial_state = BeamSearchOptimizationDecoderState(
            cell_state=self._tiled_initial_cell_state,
            previous_logits=tf.zeros([self.batch_size, self._beam_width, self._output_size], dtype=tf.float32),
            previous_score=tf.zeros([self.batch_size, self._beam_width], dtype=tf.float32),
            # During the first time step we only consider the initial beam
            num_available_beams=self._initial_num_available_beams,
            gold_beam_id=tf.zeros([self.batch_size], dtype=tf.int32),
            finished=finished)

        return (finished, start_inputs, initial_state)

def zero_state(self, batch_size, dtype=tf.float32):
        return self._cell.zero_state(batch_size, dtype)

def zero_state(self, batch_size, dtype=tf.float32):
        zeros = tf.zeros((batch_size, self._num_cells), dtype=dtype)
        return LSTMStateTuple(zeros, zeros)

def add_decoder_op(self, enc_final_state, enc_hidden_states, output_embed_matrix, training):
        cell_dec = tf.contrib.rnn.MultiRNNCell([self.make_rnn_cell(i, True) for i in range(self.config.rnn_layers)])

        encoder_hidden_size = int(enc_hidden_states.get_shape()[-1])
        decoder_hidden_size = int(cell_dec.output_size)

        # if encoder and decoder have different sizes, add a projection layer
        if encoder_hidden_size != decoder_hidden_size:
            assert False, (encoder_hidden_size, decoder_hidden_size)
            with tf.variable_scope('hidden_projection'):
                kernel = tf.get_variable('kernel', (encoder_hidden_size, decoder_hidden_size), dtype=tf.float32)

                # apply a relu to the projection for good measure
                enc_final_state = nest.map_structure(lambda x: tf.nn.relu(tf.matmul(x, kernel)), enc_final_state)
                enc_hidden_states = tf.nn.relu(tf.tensordot(enc_hidden_states, kernel, [[2], [1]]))
        else:
            # flatten and repack the state
            enc_final_state = nest.pack_sequence_as(cell_dec.state_size, nest.flatten(enc_final_state))

        if self.config.connect_output_decoder:
            cell_dec = ParentFeedingCellWrapper(cell_dec, enc_final_state)
        else:
            cell_dec = InputIgnoringCellWrapper(cell_dec, enc_final_state)
        if self.config.apply_attention:
            attention = LuongAttention(self.config.decoder_hidden_size, enc_hidden_states, self.input_length_placeholder,
                                       probability_fn=tf.nn.softmax)
            cell_dec = AttentionWrapper(cell_dec, attention,
                                        cell_input_fn=lambda inputs, _: inputs,
                                        attention_layer_size=self.config.decoder_hidden_size,
                                        initial_cell_state=enc_final_state)
            enc_final_state = cell_dec.zero_state(self.batch_size, dtype=tf.float32)
        decoder = Seq2SeqDecoder(self.config, self.input_placeholder, self.input_length_placeholder,
                                 self.output_placeholder, self.output_length_placeholder, self.batch_number_placeholder)
        return decoder.decode(cell_dec, enc_final_state, self.config.grammar.output_size, output_embed_matrix, training)

def center_loss(features, label, alfa, nrof_classes):
    """Center loss based on the paper "A Discriminative Feature Learning Approach for Deep Face Recognition"
       (http://ydwen.github.io/papers/WenECCV16.pdf)
    """
    nrof_features = features.get_shape()[1]
    centers = tf.get_variable('centers', [nrof_classes, nrof_features], dtype=tf.float32,
        initializer=tf.constant_initializer(0), trainable=False)
    label = tf.reshape(label, [-1])
    centers_batch = tf.gather(centers, label)
    diff = (1 - alfa) * (centers_batch - features)
    centers = tf.scatter_sub(centers, label, diff)
    loss = tf.reduce_mean(tf.square(features - centers_batch))
    return loss, centers

def get_batch(image_data, batch_size, batch_index):
    nrof_examples = np.size(image_data, 0)
    j = batch_index*batch_size % nrof_examples
    if j+batch_size<=nrof_examples:
        batch = image_data[j:j+batch_size,:,:,:]
    else:
        x1 = image_data[j:nrof_examples,:,:,:]
        x2 = image_data[0:nrof_examples-j,:,:,:]
        batch = np.vstack([x1,x2])
    batch_float = batch.astype(np.float32)
    return batch_float

def build_model(self):
    self.x = tf.placeholder(tf.float32, [self.reader.vocab_size], name="input")
    self.x_idx = tf.placeholder(tf.int32, [None], name="x_idx")

    self.build_encoder()
    self.build_generator()

    # Kullback Leibler divergence
    self.e_loss = -0.5 * tf.reduce_sum(1 + self.log_sigma_sq - tf.square(self.mu) - tf.exp(self.log_sigma_sq))

    # Log likelihood
    self.g_loss = -tf.reduce_sum(tf.log(tf.gather(self.p_x_i, self.x_idx) + 1e-10))

    self.loss = self.e_loss + self.g_loss

    self.encoder_var_list, self.generator_var_list = [], []
    for var in tf.trainable_variables():
      if "encoder" in var.name:
        self.encoder_var_list.append(var)
      elif "generator" in var.name:
        self.generator_var_list.append(var)

    # optimizer for alternative update
    self.optim_e = tf.train.AdamOptimizer(learning_rate=self.lr) \
                         .minimize(self.e_loss, global_step=self.step, var_list=self.encoder_var_list)
    self.optim_g = tf.train.AdamOptimizer(learning_rate=self.lr) \
                         .minimize(self.g_loss, global_step=self.step, var_list=self.generator_var_list)

    # optimizer for one shot update
    self.optim = tf.train.AdamOptimizer(learning_rate=self.lr) \
                         .minimize(self.loss, global_step=self.step)

    _ = tf.scalar_summary("encoder loss", self.e_loss)
    _ = tf.scalar_summary("generator loss", self.g_loss)
    _ = tf.scalar_summary("total loss", self.loss)

def tf_xavier_init(fan_in, fan_out, const=1.0, dtype=np.float32):
    k = const * np.sqrt(6.0 / (fan_in + fan_out))
    return tf.random_uniform((fan_in, fan_out), minval=-k, maxval=k, dtype=dtype)

def sample_gaussian(x, sigma):
    return x + tf.random_normal(tf.shape(x), mean=0.0, stddev=sigma, dtype=tf.float32)