Python tensorflow.contrib.slim.fully_connected() Examples

def encoder(self, images, is_training):
        activation_fn = leaky_relu  # tf.nn.relu
        weight_decay = 0.0
        with tf.variable_scope('encoder'):
            with slim.arg_scope([slim.batch_norm],
                                is_training=is_training):
                with slim.arg_scope([slim.conv2d, slim.fully_connected],
                                    weights_initializer=tf.truncated_normal_initializer(stddev=0.1),
                                    weights_regularizer=slim.l2_regularizer(weight_decay),
                                    normalizer_fn=slim.batch_norm,
                                    normalizer_params=self.batch_norm_params):
                    net = slim.conv2d(images, 32, [4, 4], 2, activation_fn=activation_fn, scope='Conv2d_1')
                    net = slim.conv2d(net, 64, [4, 4], 2, activation_fn=activation_fn, scope='Conv2d_2')
                    net = slim.conv2d(net, 128, [4, 4], 2, activation_fn=activation_fn, scope='Conv2d_3')
                    net = slim.conv2d(net, 256, [4, 4], 2, activation_fn=activation_fn, scope='Conv2d_4')
                    net = slim.conv2d(net, 512, [4, 4], 2, activation_fn=activation_fn, scope='Conv2d_5')
                    net = slim.flatten(net)
                    fc1 = slim.fully_connected(net, self.latent_variable_dim, activation_fn=None, normalizer_fn=None, scope='Fc_1')
                    fc2 = slim.fully_connected(net, self.latent_variable_dim, activation_fn=None, normalizer_fn=None, scope='Fc_2')
        return fc1, fc2 

def create_model(self, model_input, vocab_size, num_mixtures=None,
                   l2_penalty=1e-8, sub_scope="", original_input=None, **unused_params):
    num_supports = FLAGS.num_supports
    num_layers = FLAGS.hidden_chain_layers
    relu_cells = FLAGS.hidden_chain_relu_cells

    next_input = model_input
    support_predictions = []
    for layer in xrange(num_layers):
      sub_relu = slim.fully_connected(
          next_input,
          relu_cells,
          activation_fn=tf.nn.relu,
          weights_regularizer=slim.l2_regularizer(l2_penalty),
          scope=sub_scope+"relu-%d"%layer)
      sub_prediction = self.sub_model(sub_relu, vocab_size, sub_scope=sub_scope+"prediction-%d"%layer)
      relu_norm = tf.nn.l2_normalize(sub_relu, dim=1)
      next_input = tf.concat([next_input, relu_norm], axis=1)
      support_predictions.append(sub_prediction)
    main_predictions = self.sub_model(next_input, vocab_size, sub_scope=sub_scope+"-main")
    support_predictions = tf.concat(support_predictions, axis=1)
    return {"predictions": main_predictions, "support_predictions": support_predictions} 

def inference(images, keep_probability, phase_train=True, 
              bottleneck_layer_size=128, weight_decay=0.0, reuse=None):
    batch_norm_params = {
        # Decay for the moving averages.
        'decay': 0.995,
        # epsilon to prevent 0s in variance.
        'epsilon': 0.001,
        # force in-place updates of mean and variance estimates
        'updates_collections': None,
        # Moving averages ends up in the trainable variables collection
        'variables_collections': [ tf.GraphKeys.TRAINABLE_VARIABLES ],
    }
    
    with slim.arg_scope([slim.conv2d, slim.fully_connected],
                        weights_initializer=slim.initializers.xavier_initializer(), 
                        weights_regularizer=slim.l2_regularizer(weight_decay),
                        normalizer_fn=slim.batch_norm,
                        normalizer_params=batch_norm_params):
        return inception_resnet_v1(images, is_training=phase_train,
              dropout_keep_prob=keep_probability, bottleneck_layer_size=bottleneck_layer_size, reuse=reuse) 

def inference(images, keep_probability, phase_train=True,  # @UnusedVariable
              bottleneck_layer_size=128, bottleneck_layer_activation=None, weight_decay=0.0, reuse=None):  # @UnusedVariable
    batch_norm_params = {
        # Decay for the moving averages.
        'decay': 0.995,
        # epsilon to prevent 0s in variance.
        'epsilon': 0.001,
        # force in-place updates of mean and variance estimates
        'updates_collections': None,
        # Moving averages ends up in the trainable variables collection
        'variables_collections': [ tf.GraphKeys.TRAINABLE_VARIABLES ],
    }
    
    with slim.arg_scope([slim.conv2d, slim.fully_connected],
                        weights_initializer=tf.truncated_normal_initializer(stddev=0.1),
                        weights_regularizer=slim.l2_regularizer(weight_decay),
                        normalizer_fn=slim.batch_norm,
                        normalizer_params=batch_norm_params):
        size = np.prod(images.get_shape()[1:].as_list())
        net = slim.fully_connected(tf.reshape(images, (-1,size)), bottleneck_layer_size, activation_fn=None, 
                scope='Bottleneck', reuse=False)
        return net, None 

def mobilenetv2_scope(is_training=True,
                      trainable=True,
                      weight_decay=0.00004,
                      stddev=0.09,
                      dropout_keep_prob=0.8,
                      bn_decay=0.997):
  """Defines Mobilenet training scope.
  In default. We do not use BN

  ReWrite the scope.
  """
  batch_norm_params = {
      'is_training': False,
      'trainable': False,
      'decay': bn_decay,
  }
  with slim.arg_scope(training_scope(is_training=is_training, weight_decay=weight_decay)):
      with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.separable_conv2d],
                          trainable=trainable):
          with slim.arg_scope([slim.batch_norm], **batch_norm_params) as sc:
              return sc 

def encoder(self, images, is_training):
        activation_fn = leaky_relu  # tf.nn.relu
        weight_decay = 0.0
        with tf.variable_scope('encoder'):
            with slim.arg_scope([slim.batch_norm],
                                is_training=is_training):
                with slim.arg_scope([slim.conv2d, slim.fully_connected],
                                    weights_initializer=tf.truncated_normal_initializer(stddev=0.1),
                                    weights_regularizer=slim.l2_regularizer(weight_decay),
                                    normalizer_fn=slim.batch_norm,
                                    normalizer_params=self.batch_norm_params):
                    net = slim.conv2d(images, 32, [4, 4], 2, activation_fn=activation_fn, scope='Conv2d_1')
                    net = slim.conv2d(net, 64, [4, 4], 2, activation_fn=activation_fn, scope='Conv2d_2')
                    net = slim.conv2d(net, 128, [4, 4], 2, activation_fn=activation_fn, scope='Conv2d_3')
                    net = slim.conv2d(net, 256, [4, 4], 2, activation_fn=activation_fn, scope='Conv2d_4')
                    net = slim.flatten(net)
                    fc1 = slim.fully_connected(net, self.latent_variable_dim, activation_fn=None, normalizer_fn=None, scope='Fc_1')
                    fc2 = slim.fully_connected(net, self.latent_variable_dim, activation_fn=None, normalizer_fn=None, scope='Fc_2')
        return fc1, fc2 

def E(self, images, is_training = False, reuse=False):
	
	if images.get_shape()[3] == 3:
	    images = tf.image.rgb_to_grayscale(images)
	
	with tf.variable_scope('encoder',reuse=reuse):
	    with slim.arg_scope([slim.fully_connected], activation_fn=tf.nn.relu):
		with slim.arg_scope([slim.conv2d], activation_fn=tf.nn.relu, padding='VALID'):
		    net = slim.conv2d(images, 64, 5, scope='conv1')
		    net = slim.max_pool2d(net, 2, stride=2, scope='pool1')
		    net = slim.conv2d(net, 128, 5, scope='conv2')
		    net = slim.max_pool2d(net, 2, stride=2, scope='pool2')
		    net = tf.contrib.layers.flatten(net)
		    net = slim.fully_connected(net, 1024, activation_fn=tf.nn.relu, scope='fc3')
		    net = slim.dropout(net, 0.5, is_training=is_training)
		    net = slim.fully_connected(net, self.hidden_repr_size, activation_fn=tf.tanh,scope='fc4')
		    # dropout here or not?
		    #~ net = slim.dropout(net, 0.5, is_training=is_training)
		    return net 

def _build_aux_head(net, end_points, num_classes, hparams, scope):
  """Auxiliary head used for all models across all datasets."""
  with tf.variable_scope(scope):
    aux_logits = tf.identity(net)
    with tf.variable_scope('aux_logits'):
      aux_logits = slim.avg_pool2d(
          aux_logits, [5, 5], stride=3, padding='VALID')
      aux_logits = slim.conv2d(aux_logits, 128, [1, 1], scope='proj')
      aux_logits = slim.batch_norm(aux_logits, scope='aux_bn0')
      aux_logits = tf.nn.relu(aux_logits)
      # Shape of feature map before the final layer.
      shape = aux_logits.shape
      if hparams.data_format == 'NHWC':
        shape = shape[1:3]
      else:
        shape = shape[2:4]
      aux_logits = slim.conv2d(aux_logits, 768, shape, padding='VALID')
      aux_logits = slim.batch_norm(aux_logits, scope='aux_bn1')
      aux_logits = tf.nn.relu(aux_logits)
      aux_logits = contrib_layers.flatten(aux_logits)
      aux_logits = slim.fully_connected(aux_logits, num_classes)
      end_points['AuxLogits'] = aux_logits 

def create_model(self, model_input, vocab_size, num_mixtures=None,
                   l2_penalty=1e-8, sub_scope="", original_input=None, **unused_params):
    num_supports = FLAGS.num_supports
    num_layers = FLAGS.hidden_chain_layers
    relu_cells = FLAGS.hidden_chain_relu_cells

    next_input = model_input
    support_predictions = []
    for layer in xrange(num_layers):
      sub_relu = slim.fully_connected(
          next_input,
          relu_cells,
          activation_fn=tf.nn.relu,
          weights_regularizer=slim.l2_regularizer(l2_penalty),
          scope=sub_scope+"relu-%d"%layer)
      sub_prediction = self.sub_model(sub_relu, vocab_size, sub_scope=sub_scope+"prediction-%d"%layer)
      relu_norm = tf.nn.l2_normalize(sub_relu, dim=1)
      next_input = tf.concat([model_input, relu_norm], axis=1)
      support_predictions.append(sub_prediction)
    main_predictions = self.sub_model(next_input, vocab_size, sub_scope=sub_scope+"-main")
    support_predictions = tf.concat(support_predictions, axis=1)
    return {"predictions": main_predictions, "support_predictions": support_predictions} 

def _arg_scope(self, is_training, reuse=None):
        weight_decay = 0.0
        keep_probability = 1.0

        batch_norm_params = {
            'is_training': is_training,
            # Decay for the moving averages.
            'decay': 0.995,
            # epsilon to prevent 0s in variance.
            'epsilon': 0.001
        }

        with slim.arg_scope([slim.conv2d, slim.fully_connected],
                            weights_initializer=slim.xavier_initializer_conv2d(uniform=True),
                            weights_regularizer=slim.l2_regularizer(weight_decay),
                            normalizer_fn=slim.batch_norm,
                            normalizer_params=batch_norm_params):
            with tf.variable_scope(self._scope, self._scope, reuse=reuse):
                with slim.arg_scope([slim.batch_norm, slim.dropout],
                                    is_training=is_training) as sc:
                    return sc 

def _extra_conv_arg_scope(weight_decay=0.00001, activation_fn=None, normalizer_fn=None):

  with slim.arg_scope(
      [slim.conv2d, slim.conv2d_transpose],
      padding='SAME',
      weights_regularizer=slim.l2_regularizer(weight_decay),
      weights_initializer=tf.truncated_normal_initializer(stddev=0.001),
      activation_fn=activation_fn,
      normalizer_fn=normalizer_fn,) as arg_sc:
    with slim.arg_scope(
      [slim.fully_connected],
          weights_regularizer=slim.l2_regularizer(weight_decay),
          weights_initializer=tf.truncated_normal_initializer(stddev=0.001),
          activation_fn=activation_fn,
          normalizer_fn=normalizer_fn) as arg_sc:
          return arg_sc 

def mobilenetv2_scope(is_training=True,
                      trainable=True,
                      weight_decay=0.00004,
                      stddev=0.09,
                      dropout_keep_prob=0.8,
                      bn_decay=0.997):
  """Defines Mobilenet training scope.
  In default. We do not use BN

  ReWrite the scope.
  """
  batch_norm_params = {
      'is_training': False,
      'trainable': False,
      'decay': bn_decay,
  }
  with slim.arg_scope(training_scope(is_training=is_training, weight_decay=weight_decay)):
      with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.separable_conv2d],
                          trainable=trainable):
          with slim.arg_scope([slim.batch_norm], **batch_norm_params) as sc:
              return sc 

def squeeze_excitation_layer(input_x, out_dim, ratio, layer_name, is_training):
    with tf.name_scope(layer_name):
        # Global_Average_Pooling
        squeeze = tf.reduce_mean(input_x, [1, 2])

        excitation = slim.fully_connected(inputs=squeeze,
                                          num_outputs=out_dim // ratio,
                                          weights_initializer=cfgs.BBOX_INITIALIZER,
                                          activation_fn=tf.nn.relu,
                                          trainable=is_training,
                                          scope=layer_name+'_fully_connected1')

        excitation = slim.fully_connected(inputs=excitation,
                                          num_outputs=out_dim,
                                          weights_initializer=cfgs.BBOX_INITIALIZER,
                                          activation_fn=tf.nn.sigmoid,
                                          trainable=is_training,
                                          scope=layer_name + '_fully_connected2')

        excitation = tf.reshape(excitation, [-1, 1, 1, out_dim])

        # scale = input_x * excitation

        return excitation 

def mobilenetv2_scope(is_training=True,
                      trainable=True,
                      weight_decay=0.00004,
                      stddev=0.09,
                      dropout_keep_prob=0.8,
                      bn_decay=0.997):
  """Defines Mobilenet training scope.
  In default. We do not use BN
  ReWrite the scope.
  """
  batch_norm_params = {
      'is_training': False,
      'trainable': False,
      'decay': bn_decay,
  }
  with slim.arg_scope(training_scope(is_training=is_training, weight_decay=weight_decay)):
      with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.separable_conv2d],
                          trainable=trainable):
          with slim.arg_scope([slim.batch_norm], **batch_norm_params) as sc:
              return sc 

def mobilenetv2_scope(is_training=True,
                      trainable=True,
                      weight_decay=0.00004,
                      stddev=0.09,
                      dropout_keep_prob=0.8,
                      bn_decay=0.997):
  """Defines Mobilenet training scope.
  In default. We do not use BN
  ReWrite the scope.
  """
  batch_norm_params = {
      'is_training': False,
      'trainable': False,
      'decay': bn_decay,
  }
  with slim.arg_scope(training_scope(is_training=is_training, weight_decay=weight_decay)):
      with slim.arg_scope([slim.conv2d, slim.fully_connected, slim.separable_conv2d],
                          trainable=trainable):
          with slim.arg_scope([slim.batch_norm], **batch_norm_params) as sc:
              return sc 

def create_model(self,
                   model_input,
                   vocab_size,
                   num_mixtures=None,
                   l2_penalty=1e-8,
                   sub_scope="",
                   original_input=None, 
                   **unused_params):

    num_methods = model_input.get_shape().as_list()[-1]
    num_features = model_input.get_shape().as_list()[-2]

    original_input = tf.nn.l2_normalize(original_input, dim=1)
    gate_activations = slim.fully_connected(
        original_input,
        num_methods,
        activation_fn=tf.nn.softmax,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="gates"+sub_scope)

    output = tf.einsum("ijk,ik->ij", model_input, gate_activations)
    return {"predictions": output} 

def inference(images, keep_probability, phase_train=True, 
              bottleneck_layer_size=128, weight_decay=0.0, reuse=None):
    batch_norm_params = {
        # Decay for the moving averages.
        'decay': 0.995,
        # epsilon to prevent 0s in variance.
        'epsilon': 0.001,
        # force in-place updates of mean and variance estimates
        'updates_collections': None,
        # Moving averages ends up in the trainable variables collection
        'variables_collections': [ tf.GraphKeys.TRAINABLE_VARIABLES ],
}
    with slim.arg_scope([slim.conv2d, slim.fully_connected],
                        weights_initializer=slim.initializers.xavier_initializer(), 
                        weights_regularizer=slim.l2_regularizer(weight_decay),
                        normalizer_fn=slim.batch_norm,
                        normalizer_params=batch_norm_params):
        return inception_resnet_v2(images, is_training=phase_train,
              dropout_keep_prob=keep_probability, bottleneck_layer_size=bottleneck_layer_size, reuse=reuse) 

def sub_model(self, model_input, vocab_size, num_mixtures=None, 
                l2_penalty=1e-8, sub_scope="", 
                dropout=False, keep_prob=None, noise_level=None,
                **unused_params):
    num_mixtures = num_mixtures or FLAGS.moe_num_mixtures

    if dropout:
      model_input = tf.nn.dropout(model_input, keep_prob=keep_prob)

    gate_activations = slim.fully_connected(
        model_input,
        vocab_size * (num_mixtures + 1),
        activation_fn=None,
        biases_initializer=None,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="gates-"+sub_scope)
    expert_activations = slim.fully_connected(
        model_input,
        vocab_size * num_mixtures,
        activation_fn=None,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="experts-"+sub_scope)

    gating_distribution = tf.nn.softmax(tf.reshape(
        gate_activations,
        [-1, num_mixtures + 1]))  # (Batch * #Labels) x (num_mixtures + 1)
    expert_distribution = tf.nn.sigmoid(tf.reshape(
        expert_activations,
        [-1, num_mixtures]))  # (Batch * #Labels) x num_mixtures

    final_probabilities_by_class_and_batch = tf.reduce_sum(
        gating_distribution[:, :num_mixtures] * expert_distribution, 1)
    final_probabilities = tf.reshape(final_probabilities_by_class_and_batch,
                                     [-1, vocab_size])
    return final_probabilities 

def create_model(self, model_input, vocab_size, num_mixtures=None,
                   l2_penalty=1e-8, sub_scope="", original_input=None, 
                   dropout=False, keep_prob=None, noise_level=None,
                   num_frames=None,
                   **unused_params):

    num_supports = FLAGS.num_supports
    num_layers = FLAGS.deep_chain_layers
    relu_cells = FLAGS.deep_chain_relu_cells
    relu_type = FLAGS.deep_chain_relu_type
    use_length = FLAGS.deep_chain_use_length

    next_input = model_input
    support_predictions = []
    for layer in xrange(num_layers):
      sub_prediction = self.sub_model(next_input, vocab_size, sub_scope=sub_scope+"prediction-%d"%layer, dropout=dropout, keep_prob=keep_prob, noise_level=noise_level)
      sub_activation = slim.fully_connected(
          sub_prediction,
          relu_cells,
          activation_fn=None,
          weights_regularizer=slim.l2_regularizer(l2_penalty),
          scope=sub_scope+"relu-%d"%layer)

      if relu_type == "elu":
        sub_relu = tf.nn.elu(sub_activation)
      else: # default: relu
        sub_relu = tf.nn.relu(sub_activation)

      if noise_level is not None:
        print "adding noise to sub_relu, level = ", noise_level
        sub_relu = sub_relu + tf.random_normal(tf.shape(sub_relu), mean=0.0, stddev=noise_level)

      relu_norm = tf.nn.l2_normalize(sub_relu, dim=1)
      next_input = tf.concat([next_input, relu_norm], axis=1)
      support_predictions.append(sub_prediction)
    main_predictions = self.sub_model(next_input, vocab_size, sub_scope=sub_scope+"-main")
    support_predictions = tf.concat(support_predictions, axis=1)
    return {"predictions": main_predictions, "support_predictions": support_predictions} 

def create_model(self, model_input, vocab_size, l2_penalty=1e-8, **unused_params):
    """Creates a logistic model.

    Args:
      model_input: 'batch' x 'num_features' matrix of input features.
      vocab_size: The number of classes in the dataset.

    Returns:
      A dictionary with a tensor containing the probability predictions of the
      model in the 'predictions' key. The dimensions of the tensor are
      batch_size x num_classes."""
    output = slim.fully_connected(
        model_input, vocab_size, activation_fn=tf.nn.sigmoid,
        weights_regularizer=slim.l2_regularizer(l2_penalty))
    return {"predictions": output} 

def sub_model(self, model_input, vocab_size, num_mixtures=None, 
                l2_penalty=1e-8, sub_scope="", 
                dropout=False, keep_prob=None, noise_level=None,
                **unused_params):
    num_mixtures = num_mixtures or FLAGS.moe_num_mixtures

    if dropout:
      model_input = tf.nn.dropout(model_input, keep_prob=keep_prob)

    gate_activations = slim.fully_connected(
        model_input,
        vocab_size * (num_mixtures + 1),
        activation_fn=None,
        biases_initializer=None,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="gates-"+sub_scope)
    expert_activations = slim.fully_connected(
        model_input,
        vocab_size * num_mixtures,
        activation_fn=None,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="experts-"+sub_scope)

    gating_distribution = tf.nn.softmax(tf.reshape(
        gate_activations,
        [-1, num_mixtures + 1]))  # (Batch * #Labels) x (num_mixtures + 1)
    expert_distribution = tf.nn.sigmoid(tf.reshape(
        expert_activations,
        [-1, num_mixtures]))  # (Batch * #Labels) x num_mixtures

    final_probabilities_by_class_and_batch = tf.reduce_sum(
        gating_distribution[:, :num_mixtures] * expert_distribution, 1)
    final_probabilities = tf.reshape(final_probabilities_by_class_and_batch,
                                     [-1, vocab_size])
    return final_probabilities 

def create_model(self, model_input, vocab_size, num_mixtures=None,
                   l2_penalty=1e-8, sub_scope="", original_input=None, 
                   num_frames=None, **unused_params):
    num_supports = FLAGS.num_supports
    num_layers = FLAGS.deep_chain_layers
    relu_cells = FLAGS.deep_chain_relu_cells
    use_length = FLAGS.deep_chain_use_length

    if use_length:
      model_input = tf.concat([model_input, self.get_length_code(num_frames)], axis=1)

    next_input = model_input
    support_predictions = []
    for layer in xrange(num_layers):
      sub_prediction = self.sub_model(next_input, vocab_size, sub_scope=sub_scope+"prediction-%d"%layer)
      sub_relu = slim.fully_connected(
          sub_prediction,
          relu_cells,
          activation_fn=tf.nn.relu,
          weights_regularizer=slim.l2_regularizer(l2_penalty),
          scope=sub_scope+"relu-%d"%layer)
      relu_norm = tf.nn.l2_normalize(sub_relu, dim=1)
      next_input = tf.concat([model_input, relu_norm], axis=1)
      support_predictions.append(sub_prediction)
    main_predictions = self.sub_model(next_input, vocab_size, sub_scope=sub_scope+"-main")
    support_predictions = tf.concat(support_predictions, axis=1)
    return {"predictions": main_predictions, "support_predictions": support_predictions} 

def sub_model(self, model_input, vocab_size, num_mixtures=None, 
                l2_penalty=1e-8, sub_scope="", 
                dropout=False, keep_prob=None, noise_level=None,
                **unused_params):
    num_mixtures = num_mixtures or FLAGS.moe_num_mixtures

    if dropout:
      print "adding dropout to moe, keep_prob = ", keep_prob
      model_input = tf.nn.dropout(model_input, keep_prob=keep_prob)

    gate_activations = slim.fully_connected(
        model_input,
        vocab_size * (num_mixtures + 1),
        activation_fn=None,
        biases_initializer=None,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="gates-"+sub_scope)
    expert_activations = slim.fully_connected(
        model_input,
        vocab_size * num_mixtures,
        activation_fn=None,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="experts-"+sub_scope)

    gating_distribution = tf.nn.softmax(tf.reshape(
        gate_activations,
        [-1, num_mixtures + 1]))  # (Batch * #Labels) x (num_mixtures + 1)
    expert_distribution = tf.nn.sigmoid(tf.reshape(
        expert_activations,
        [-1, num_mixtures]))  # (Batch * #Labels) x num_mixtures

    final_probabilities_by_class_and_batch = tf.reduce_sum(
        gating_distribution[:, :num_mixtures] * expert_distribution, 1)
    final_probabilities = tf.reshape(final_probabilities_by_class_and_batch,
                                     [-1, vocab_size])
    return final_probabilities 

def create_model(self, model_input, vocab_size, num_mixtures=None,
                   l2_penalty=1e-8, sub_scope="", original_input=None, **unused_params):
    num_supports = FLAGS.num_supports
    input_size = model_input.shape.as_list()[1]
    support_predictions = self.sub_model(model_input, num_supports, sub_scope=sub_scope+"-support")
    main_relu = slim.fully_connected(
        model_input,
        input_size,
        activation_fn=tf.nn.relu,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="main-relu-"+sub_scope)
    main_input = tf.concat([main_relu, support_predictions], axis=1)
    main_predictions = self.sub_model(main_input, vocab_size, sub_scope=sub_scope+"-main")
    return {"predictions": main_predictions, "support_predictions": support_predictions} 

def create_model(self, model_input, vocab_size, l2_penalty=1e-8, original_input=None, **unused_params):
    """Creates a logistic model.

    Args:
      model_input: 'batch' x 'num_features' matrix of input features.
      vocab_size: The number of classes in the dataset.

    Returns:
      A dictionary with a tensor containing the probability predictions of the
      model in the 'predictions' key. The dimensions of the tensor are
      batch_size x num_classes."""
    output = slim.fully_connected(
        model_input, vocab_size, activation_fn=tf.nn.sigmoid,
        weights_regularizer=slim.l2_regularizer(l2_penalty))
    return {"predictions": output} 

def create_model(self, model_input, vocab_size, l2_penalty=1e-8, **unused_params):
        """Creates a logistic model.

        Args:
          model_input: 'batch' x 'num_features' matrix of input features.
          vocab_size: The number of classes in the dataset.

        Returns:
          A dictionary with a tensor containing the probability predictions of the
          model in the 'predictions' key. The dimensions of the tensor are
          batch_size x num_classes."""
        model_input = tf.cast(model_input,dtype=tf.float32)
        hidden_size = FLAGS.hidden_size

        model_mask, indices_input = tf.nn.top_k(model_input, k=FLAGS.top_k)
        indices_input = tf.reshape(indices_input, [-1])
        models_mask = tf.reshape(model_mask, [-1,FLAGS.top_k,1])
        with tf.name_scope("embedding"):
            embeddings = tf.Variable(
                tf.random_uniform([vocab_size, hidden_size], -1.0, 1.0))
            embed = tf.nn.embedding_lookup(embeddings, indices_input)
            output = slim.fully_connected(
                embed,
                vocab_size,
                activation_fn=tf.nn.sigmoid,
                weights_regularizer=slim.l2_regularizer(l2_penalty),
                scope="output")
        indices_one_hot = tf.one_hot(indices_input, vocab_size)
        output = output * (1 - indices_one_hot) + indices_one_hot
        output_val = tf.reshape(output,[-1,FLAGS.top_k,vocab_size])
        predictions_val = tf.reduce_sum(output_val*models_mask, axis=1)/tf.reduce_sum(models_mask, axis=1)
        return {"predictions": output, "predictions_val": predictions_val} 

def relu(self, model_input, relu_cells, 
               l2_penalty=1e-8, sub_scope=""):
    sub_activation = slim.fully_connected(
        model_input,
        relu_cells,
        activation_fn=tf.nn.relu,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="relu-"+sub_scope)
    return sub_activation 

def create_model(self,
                   model_input,
                   vocab_size,
                   num_mixtures=None,
                   l2_penalty=1e-8,
                   sub_scope="",
                   original_input=None, 
                   **unused_params):

    num_relu = FLAGS.attention_relu_cells
    num_methods = model_input.get_shape().as_list()[-1]
    num_features = model_input.get_shape().as_list()[-2]

    original_input = tf.nn.l2_normalize(original_input, dim=1)
    model_input_list = tf.unstack(model_input, axis=2)
    
    relu_units = [self.relu(original_input, num_relu, sub_scope="input")]
    i = 0
    for mi in model_input_list:
      relu_units.append(self.relu(mi, num_relu, sub_scope="sub"+str(i)))
      i += 1

    gate_activations = slim.fully_connected(
        tf.concat(relu_units, axis=1),
        num_methods,
        activation_fn=None,
        biases_initializer=None,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="gate")
    gate = tf.nn.softmax(gate_activations)
    output = tf.einsum("ijk,ik->ij", model_input, gate)
    return {"predictions": output} 

def create_model(self,
                   model_input,
                   vocab_size,
                   num_mixtures=None,
                   l2_penalty=1e-8,
                   sub_scope="",
                   original_input=None, 
                   **unused_params):

    num_methods = model_input.get_shape().as_list()[-1]
    num_features = model_input.get_shape().as_list()[-2]
    num_mixtures = FLAGS.moe_num_mixtures

    # gating coefficients
    original_input = tf.nn.l2_normalize(original_input, dim=1)
    mean_output = tf.reduce_mean(model_input, axis=2)
    ## batch_size x moe_num_mixtures
    gate_activations = slim.fully_connected(
        tf.concat([original_input, mean_output], axis=1),
        num_mixtures,
        activation_fn=tf.nn.softmax,
        weights_regularizer=slim.l2_regularizer(l2_penalty),
        scope="gates"+sub_scope)

    # matrix
    weight_var = tf.get_variable("ensemble_weight",
        shape=[num_mixtures, num_methods],
        regularizer=slim.l2_regularizer(l2_penalty))

    # weight
    gated_weight = tf.einsum("ij,jk->ik", gate_activations, weight_var)
    weight = tf.nn.softmax(gated_weight)
    
    # weighted output
    output = tf.einsum("ik,ijk->ij", weight, model_input)
    return {"predictions": output} 

def create_model(self, model_input, vocab_size, l2_penalty=1e-8, original_input=None, **unused_params):
    """Creates a logistic model.

    Args:
      model_input: 'batch' x 'num_features' matrix of input features.
      vocab_size: The number of classes in the dataset.

    Returns:
      A dictionary with a tensor containing the probability predictions of the
      model in the 'predictions' key. The dimensions of the tensor are
      batch_size x num_classes."""
    output = slim.fully_connected(
        model_input, vocab_size, activation_fn=tf.nn.sigmoid,
        weights_regularizer=slim.l2_regularizer(l2_penalty))
    return {"predictions": output}