Python tensorflow.AUTO_REUSE属性代码示例

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def fprop(self, x, **kwargs):
        del kwargs
        my_conv = functools.partial(tf.layers.conv2d,
                                    kernel_size=3,
                                    strides=2,
                                    padding='valid',
                                    activation=tf.nn.relu,
                                    kernel_initializer=HeReLuNormalInitializer)
        my_dense = functools.partial(
            tf.layers.dense, kernel_initializer=HeReLuNormalInitializer)

        with tf.variable_scope(self.scope, reuse=tf.AUTO_REUSE):
            for depth in [96, 256, 384, 384, 256]:
                x = my_conv(x, depth)
            y = tf.layers.flatten(x)
            y = my_dense(y, 4096, tf.nn.relu)
            y = fc7 = my_dense(y, 4096, tf.nn.relu)
            y = my_dense(y, 1000)
            return {'fc7': fc7,
                    self.O_LOGITS: y,
                    self.O_PROBS: tf.nn.softmax(logits=y)} 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def build_critic(x, h, output_size, scope, n_layers, size, gru_size, recurrent=True, activation=tf.tanh, output_activation=None, regularizer=None):
    """
    build recurrent critic

    arguments:
        regularizer: regularization for weights
        (see `build_policy()` for rest)

    n.b. the policy and critic should not share weights
    """
    with tf.variable_scope(scope, reuse=tf.AUTO_REUSE):
        if recurrent:
            x, h = build_rnn(x, h, gru_size, scope, n_layers, size, activation=activation, output_activation=output_activation, regularizer=regularizer)
        else:
            x = tf.reshape(x, (-1, x.get_shape()[1]*x.get_shape()[2]))
            x = build_mlp(x, gru_size, scope, n_layers + 1, size, activation=activation, output_activation=activation, regularizer=regularizer)
        x = tf.layers.dense(x, output_size, activation=output_activation, name='decoder', kernel_regularizer=regularizer, bias_regularizer=regularizer)
    return x 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def reward_prediction(
      self, input_image, input_reward, action, lstm_state, latent):
    """Builds a reward prediction network."""
    conv_size = self.tinyify([32, 32, 16, 4])
    lstm_size = self.tinyify([32, 64, 128, 64, 32])

    with tf.variable_scope("reward_pred", reuse=tf.AUTO_REUSE):
      hidden5, _ = self.bottom_part_tower(
          input_image, input_reward, action, latent,
          lstm_state, lstm_size, conv_size)

      x = hidden5
      x = slim.batch_norm(x, scope="reward_bn0")
      x = slim.conv2d(x, conv_size[1], [3, 3], scope="reward_conv1")
      x = slim.batch_norm(x, scope="reward_bn1")
      x = slim.conv2d(x, conv_size[2], [3, 3], scope="reward_conv2")
      x = slim.batch_norm(x, scope="reward_bn2")
      x = slim.conv2d(x, conv_size[3], [3, 3], scope="reward_conv3")

      pred_reward = self.decode_to_shape(
          x, input_reward.shape, "reward_dec")

      return pred_reward, lstm_state 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def targets_bottom(self, x, summary_prefix="targets_bottom"):  # pylint: disable=arguments-differ
    inputs = x
    with tf.variable_scope(self.name, reuse=tf.AUTO_REUSE):
      common_layers.summarize_video(inputs, summary_prefix)
      inputs_shape = common_layers.shape_list(inputs)
      # We embed each of 256=self.top_dimensionality possible pixel values.
      embedding_var = tf.get_variable(
          "pixel_embedding",
          [self.top_dimensionality, self.PIXEL_EMBEDDING_SIZE])
      hot_inputs = tf.one_hot(tf.to_int32(inputs), self.top_dimensionality)
      hot_inputs = tf.reshape(hot_inputs, [-1, self.top_dimensionality])
      embedded = tf.matmul(hot_inputs, embedding_var)
      # Let's now merge all channels that were embedded into a single vector.
      merged_size = self.PIXEL_EMBEDDING_SIZE * inputs_shape[4]
      embedded = tf.reshape(embedded, inputs_shape[:4] + [merged_size])
      transposed = common_layers.time_to_channels(embedded)
      return tf.layers.dense(
          transposed,
          self._body_input_depth,
          name="merge_pixel_embedded_frames") 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def get_vq_bottleneck(bottleneck_size, hidden_size):
  """Get lookup table for VQ bottleneck."""
  with tf.variable_scope("vq", reuse=tf.AUTO_REUSE):
    means = tf.get_variable(
        name="means",
        shape=[bottleneck_size, hidden_size],
        initializer=tf.uniform_unit_scaling_initializer())

    ema_count = tf.get_variable(
        name="ema_count",
        shape=[bottleneck_size],
        initializer=tf.constant_initializer(0),
        trainable=False)

    with tf.colocate_with(means):
      ema_means = tf.get_variable(
          name="ema_means",
          initializer=means.initialized_value(),
          trainable=False)

  return means, ema_means, ema_count 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def __init__(self, epsilon=1e-4, shape=(), scope=''):
        sess = get_session()

        self._new_mean = tf.placeholder(shape=shape, dtype=tf.float64)
        self._new_var = tf.placeholder(shape=shape, dtype=tf.float64)
        self._new_count = tf.placeholder(shape=(), dtype=tf.float64)

        
        with tf.variable_scope(scope, reuse=tf.AUTO_REUSE):
            self._mean  = tf.get_variable('mean',  initializer=np.zeros(shape, 'float64'),      dtype=tf.float64)
            self._var   = tf.get_variable('std',   initializer=np.ones(shape, 'float64'),       dtype=tf.float64)    
            self._count = tf.get_variable('count', initializer=np.full((), epsilon, 'float64'), dtype=tf.float64)

        self.update_ops = tf.group([
            self._var.assign(self._new_var),
            self._mean.assign(self._new_mean),
            self._count.assign(self._new_count)
        ])

        sess.run(tf.variables_initializer([self._mean, self._var, self._count]))
        self.sess = sess
        self._set_mean_var_count() 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def generator(self, inputs_condition):
        inputs = inputs_condition
        with tf.variable_scope("generator", reuse=tf.AUTO_REUSE):
            inputs1 = leaky_relu(conv2d("conv1", inputs, 64, 5, 2))#128x128x128
            inputs2 = leaky_relu(instanceNorm("in1", conv2d("conv2", inputs1, 128, 5, 2)))#64x64x256
            inputs3 = leaky_relu(instanceNorm("in2", conv2d("conv3", inputs2, 256, 5, 2)))#32x32x512
            inputs4 = leaky_relu(instanceNorm("in3", conv2d("conv4", inputs3, 512, 5, 2)))#16x16x512
            inputs5 = leaky_relu(instanceNorm("in4", conv2d("conv5", inputs4, 512, 5, 2)))#8x8x512
            inputs6 = leaky_relu(instanceNorm("in5", conv2d("conv6", inputs5, 512, 5, 2)))#4x4x512
            inputs7 = leaky_relu(instanceNorm("in6", conv2d("conv7", inputs6, 512, 5, 2)))#2x2x512
            inputs8 = leaky_relu(instanceNorm("in7", conv2d("conv8", inputs7, 512, 5, 2)))#1x1x512
            outputs1 = tf.nn.relu(tf.concat([tf.nn.dropout(instanceNorm("in9", deconv2d("dconv1", inputs8, 512, 5, 2)), 0.5), inputs7], axis=3))  # 2x2x512
            outputs2 = tf.nn.relu(tf.concat([tf.nn.dropout(instanceNorm("in10", deconv2d("dconv2", outputs1, 512, 5, 2)), 0.5), inputs6], axis=3))  # 4x4x512
            outputs3 = tf.nn.relu(tf.concat([tf.nn.dropout(instanceNorm("in11", deconv2d("dconv3", outputs2, 512, 5, 2)), 0.5), inputs5], axis=3))#8x8x512
            outputs4 = tf.nn.relu(tf.concat([instanceNorm("in12", deconv2d("dconv4", outputs3, 512, 5, 2)), inputs4], axis=3))#16x16x512
            outputs5 = tf.nn.relu(tf.concat([instanceNorm("in13", deconv2d("dconv5", outputs4, 256, 5, 2)), inputs3], axis=3))#32x32x256
            outputs6 = tf.nn.relu(tf.concat([instanceNorm("in14", deconv2d("dconv6", outputs5, 128, 5, 2)), inputs2], axis=3))#64x64x128
            outputs7 = tf.nn.relu(tf.concat([instanceNorm("in15", deconv2d("dconv7", outputs6, 64, 5, 2)), inputs1], axis=3))#128x128x64
            outputs8 = tf.nn.tanh((deconv2d("dconv8", outputs7, 3, 5, 2)))#256x256x3
            return outputs8 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def _init_pos_model(self, session):
        """Create POS Tagger model and initialize with random or load parameters in session."""
        # initilize config
        config_dict = load_config(self.model_config_path)
        config = get_config(config_dict, self.name)
        config.batch_size = 1
        config.num_steps = 1 # iterator one token per time
        model_var_scope = get_model_var_scope(self.var_scope, self.name)
        print ("NOTICE: Input POS Model Var Scope Name '%s'" % model_var_scope)
        # Check if self.model already exist
        if self.model is None:
            with tf.variable_scope(model_var_scope, tf.AUTO_REUSE):
                self.model = pos_model.POSTagger(is_training=False, config=config) # save object after is_training
        # Load Specific .data* ckpt file
        if len(glob.glob(self.ckpt_path + '.data*')) > 0: # file exist with pattern: 'pos.ckpt.data*'
            print("NOTICE: Loading model parameters from %s" % self.ckpt_path)
            all_vars = tf.global_variables()
            model_vars = [k for k in all_vars if model_var_scope in k.name.split("/")]
            tf.train.Saver(model_vars).restore(session, self.ckpt_path)
        else:
            print("NOTICE: Model not found, Try to run method: deepnlp.download(module='pos', name='%s')" % self.name)
            print("NOTICE: Created with fresh parameters.")
            session.run(tf.global_variables_initializer()) 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def _init_model(self, session):
        """Create Parser model and initialize with random or load parameters in session."""
        config_dict = load_config(self.model_config_path)
        config = get_config(config_dict, self.name)
        model_var_scope = get_model_var_scope(self.var_scope, self.name)
        print ("NOTICE: Initializing model var scope '%s'" % model_var_scope)
        # Check if self.model already exist
        if self.model is None:   # Create Graph Only once
            with tf.variable_scope(model_var_scope, reuse = tf.AUTO_REUSE):
                self.model = parse_model.NNParser(config=config)
        if len(glob.glob(self.ckpt_path + '.data*')) > 0: # file exist with pattern: 'parser.ckpt.data*'
            print("NOTICE: Loading model parameters from %s" % self.ckpt_path)
            all_vars = tf.global_variables()
            model_vars = [k for k in all_vars if model_var_scope in k.name.split("/")]   # Only Restore the Variable in Graph begin with parser/....
            tf.train.Saver(model_vars).restore(session, self.ckpt_path)
        else:
            print("NOTICE: Model not found, Try to run method: deepnlp.download('parse')")
            print("NOTICE: Created with fresh parameters.")
            session.run(tf.global_variables_initializer()) 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def extract_box_classifier_features(self, proposal_feature_maps, scope):
    """Extracts second stage box classifier features.

    Args:
      proposal_feature_maps: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, crop_height, crop_width, depth]
        representing the feature map cropped to each proposal.
      scope: A scope name.

    Returns:
      proposal_classifier_features: A 4-D float tensor with shape
        [batch_size * self.max_num_proposals, height, width, depth]
        representing box classifier features for each proposal.
    """
    with tf.variable_scope(
        scope, values=[proposal_feature_maps], reuse=tf.AUTO_REUSE):
      return self._extract_box_classifier_features(proposal_feature_maps, scope) 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def __init__(self, epsilon=1e-4, shape=(), scope=''):
        sess = get_session()

        self._new_mean = tf.placeholder(shape=shape, dtype=tf.float64)
        self._new_var = tf.placeholder(shape=shape, dtype=tf.float64)
        self._new_count = tf.placeholder(shape=(), dtype=tf.float64)


        with tf.variable_scope(scope, reuse=tf.AUTO_REUSE):
            self._mean  = tf.get_variable('mean',  initializer=np.zeros(shape, 'float64'),      dtype=tf.float64)
            self._var   = tf.get_variable('std',   initializer=np.ones(shape, 'float64'),       dtype=tf.float64)
            self._count = tf.get_variable('count', initializer=np.full((), epsilon, 'float64'), dtype=tf.float64)

        self.update_ops = tf.group([
            self._var.assign(self._new_var),
            self._mean.assign(self._new_mean),
            self._count.assign(self._new_count)
        ])

        sess.run(tf.variables_initializer([self._mean, self._var, self._count]))
        self.sess = sess
        self._set_mean_var_count() 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def __call__(self, obs, reuse=False):
        with tf.variable_scope(self.name, reuse=tf.AUTO_REUSE):
            x = self.network_builder(obs)
            # x = tf.layers.dense(x, self.nb_actions, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            # x = tf.nn.tanh(x)

            x = tf.layers.dense(x, self.hidden_layer1, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            x = tf.nn.tanh(x)
            x = tf.layers.dense(x, self.hidden_layer2, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            x = tf.nn.tanh(x)
            x = tf.layers.dense(x, self.hidden_layer3, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            x = tf.nn.tanh(x)
            x = tf.layers.dense(x, self.hidden_layer4, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            x = tf.nn.tanh(x)
            x = tf.layers.dense(x, self.nb_actions, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            '''
            in order to discrete the action, apply the trick of sharp sigmoid.
            with the following line: x<0 will be near 0, x>0 will be near 1
            then the discrete step in action choice wont affect too much of accuracy (with only small change of action value!)
            the key point of this step is to keep most part of discrete operation in the tensor graph, so as to be back propagated
            '''

            x = tf.nn.sigmoid(1000*x)  # sigmoid ~ (0,1), tanh ~ (-1 , 1)

        return x 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def __call__(self, obs, reuse=False):
        with tf.variable_scope(self.name, reuse=tf.AUTO_REUSE):
            x = self.network_builder(obs)
            # x = tf.layers.dense(x, self.nb_actions, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            # x = tf.nn.tanh(x)

            x = tf.layers.dense(x, self.hidden_layer1, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            x = tf.nn.tanh(x)
            x = tf.layers.dense(x, self.hidden_layer2, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            x = tf.nn.tanh(x)
            # x = tf.layers.dense(x, self.hidden_layer3, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            # x = tf.nn.tanh(x)
            # x = tf.layers.dense(x, self.hidden_layer4, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            # x = tf.nn.tanh(x)
            x = tf.layers.dense(x, self.nb_actions, kernel_initializer=tf.random_uniform_initializer(minval=-3e-3, maxval=3e-3))
            # x = self.step_activation(tf.nn.sigmoid(x))  # sigmoid ~ (0,1), tanh ~ (-1 , 1)
            '''1000 makes the binarization of output of sigmoid has smaller difference 
            that cannot be backpropagated with tensorflow'''
            x = tf.nn.sigmoid(1000*x)  # sigmoid ~ (0,1), tanh ~ (-1 , 1)

        return x 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def conv2d(input_, output_dim, kernel_h=3, kernel_w=None, stride_h=1, stride_w=None, padding='SAME', reuse=False, initializer=None, use_bias = True, name="conv2d"):
  
  if kernel_w == None: kernel_w = kernel_h
  if stride_w == None: stride_w = stride_h
  if initializer == None: initializer = tf.contrib.layers.xavier_initializer()

  with tf.variable_scope(name, reuse = tf.AUTO_REUSE):
    if reuse==True: scope.reuse_variables()
    w = tf.get_variable('w', [kernel_h, kernel_w, input_.get_shape()[-1], output_dim],
      initializer=initializer)
    conv = tf.nn.conv2d(input_, w, strides=[1,stride_h, stride_w, 1], padding=padding)

    if use_bias:
      b = tf.get_variable('bias', [output_dim], initializer=tf.constant_initializer(0.0))
      conv = tf.nn.bias_add(conv, b)

    return conv 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import AUTO_REUSE [as 别名]
def res_block(input_, output_dim, name='res_block', is_dropout=False, drop_p=0.5):
  shortcut = input_
  num_input_c = shortcut.shape.as_list()[-1]

  with tf.variable_scope(name, reuse=tf.AUTO_REUSE):
    conv = conv2d(input_, output_dim, name=name+'/conv1')
    conv = norm_layer(conv, 'instance')
    conv = tf.nn.relu(conv)

    if is_dropout:
      conv = tf.nn.dropout(conv, keep_prob = drop_p)

    #conv = conv2d(conv, output_dim, num_input_c * 2, name=name+'/conv2')
    conv = conv2d(conv, output_dim,  name=name+'/conv2')
    conv = norm_layer(conv, 'instance')

    conv = tf.identity(conv+shortcut, name='residual_block_output')

  return conv