BIDI_ML_INFO_EXTRACTION/BiddingKG/dl/common/models.py

'''
Created on 2019年4月22日

@author: User
'''

from keras import models,layers,losses,optimizers
from keras.callbacks import ModelCheckpoint
import numpy as np
from BiddingKG.dl.common.Utils import *
import keras.backend as K
import tensorflow as tf
import math
import six
layers.Dense

def getTextCNNModel(input_shape,vocab,embedding_weights,classes):

    def resize(x):
        _shape = shape_list(x)
        print("#$",_shape)
        x1 = tf.reshape(x,[_shape[0],2,_shape[1]//2,_shape[2]])  # type: object
        print("--")
        x2 = tf.transpose(x1,[1,0,2,3])
        x_l,x_r = tf.split(x2,[1,1],axis=0)
        return [tf.squeeze(x_l,axis=0),tf.squeeze(x_r,axis=0)]

    def resize_input(x):
        _shape = shape_list(x)
        x1 = tf.reshape(x,[_shape[0],3,_shape[1]//3])  # type: object
        x2 = tf.transpose(x1,[1,0,2])
        x_l,x_c,x_r = tf.split(x2,[1,1,1],axis=0)
        return [tf.squeeze(x_l,axis=0),tf.squeeze(x_c,axis=0),tf.squeeze(x_r,axis=0)]

    # assert len(input_shape)==3
    list_input = []
    for i in range(input_shape[0]):
        list_input.append(layers.Input(shape=(input_shape[1],),dtype=tf.int32,name="input%d"%(i)))
    print("list_input",list_input)
    list_embedding = []

    # if len(list_input)==1:
    #     list_resizeinput = layers.Lambda(resize_input)(list_input[0])
    # else:

    concat_input = layers.Lambda(lambda x:tf.concat(x,axis=-1))(list_input)
    embedding_input = [concat_input]
    # embedding_input = list_input

    embedding = layers.Embedding(len(vocab),input_shape[2],weights=[embedding_weights] if embedding_weights is not None else None,trainable=True,name="char_embeding")
    for i in range(len(embedding_input)):
        print(i)
        list_embedding.append(embedding(embedding_input[i]))
        print(list_embedding)


    #
    set_variables = set()
    for v in tf.trainable_variables():
        set_variables.add(v.name)

    list_bert = []
    for i in range(len(list_embedding)):
        for v in tf.trainable_variables():
            set_variables.add(v.name)
        bert_layer = layers.Lambda(lambda x:transformer_model(input_tensor=x,name="bert%d"%(i)),trainable=True,name="bert%d"%(i))
        list_bert.append(bert_layer(list_embedding[i]))
        #set bert_weights to trainable
        bert_weights = []
        for v in tf.trainable_variables():
            if v.name not in set_variables:
                print("++++",v.name)
                bert_weights.append(v)
        bert_layer._trainable_weights = bert_weights

    _resize = layers.Lambda(lambda x:resize(x))(list_bert[0])
    list_w2v = _resize


    list_conv = []
    list_kernel = [2,5,8]
    for i in range(len(list_w2v)):
        list_temp = []
        for kernel in list_kernel:
            list_temp.append(layers.Conv1D(10,kernel,strides=1,padding="same",activation="relu")(list_w2v[i]))
        list_conv.append(layers.merge(list_temp,mode="concat"))
    layers.Conv2D
    list_matrix = []
    for i in range(len(list_conv)):
        list_matrix.append(layers.Dense(12,activation="relu")(list_conv[i]))
        
    if len(list_matrix)>1:
        ave = layers.merge(list_matrix,mode="concat")
        
        dropout = layers.Dropout(0.3)(ave)
    else:
        dropout = layers.Dropout(0.3)(list_matrix[0])
    
    flatten = layers.Flatten()(dropout)
    
    matrix = layers.Dense(classes*10,activation="relu")(flatten)
    
    out = layers.Dense(classes,activation="softmax")(matrix)
    
    model = models.Model(list_input,out)
    
    model.compile(optimizer=optimizers.Adadelta(),loss=losses.categorical_crossentropy,metrics=[precision,recall,f1_score])
    
    model.summary()
    
    return model

class Attention(layers.Layer):
    
        def __init__(self, **kwargs):
            super(Attention, self).__init__(**kwargs)
    
            
        def build(self, input_shape):
            # W: (EMBED_SIZE, 1)
            # b: (MAX_TIMESTEPS, 1)
            # u: (MAX_TIMESTEPS, MAX_TIMESTEPS)
            self.W = self.add_weight(name="W_{:s}".format(self.name),
                                     shape=(input_shape[-1], 1),
                                     initializer="normal")
            self.b = self.add_weight(name="b_{:s}".format(self.name),
                                     shape=(input_shape[1], 1),
                                     initializer="zeros")
            self.u = self.add_weight(name="u_{:s}".format(self.name),
                                     shape=(input_shape[1], input_shape[1]),
                                     initializer="normal")
            super(Attention, self).build(input_shape)
    
    
        def call(self, x, mask=None):
            # input: (BATCH_SIZE, MAX_TIMESTEPS, EMBED_SIZE)
            # et: (BATCH_SIZE, MAX_TIMESTEPS)
            et = K.squeeze(K.tanh(K.dot(x, self.W) + self.b), axis=-1)
            # at: (BATCH_SIZE, MAX_TIMESTEPS)
            at = K.dot(et, self.u)
            at = K.exp(at)
            if mask is not None:
                at *= K.cast(mask, K.floatx())
            # ot: (BATCH_SIZE, MAX_TIMESTEPS, EMBED_SIZE)
            at /= K.cast(K.sum(at, axis=1, keepdims=True) + K.epsilon(), K.floatx())
    
            atx = K.expand_dims(at, axis=-1)
            ot = atx * x
            # output: (BATCH_SIZE, EMBED_SIZE)
            return K.sum(ot, axis=1)
        def compute_mask(self, input, input_mask=None):
            # do not pass the mask to the next layers
            return None
    
    
        def compute_output_shape(self, input_shape):
            # output shape: (BATCH_SIZE, EMBED_SIZE)
            return (input_shape[0], input_shape[-1])
        
        
        def get_config(self):
            return super(Attention, self).get_config()
        
def gelu(x):
    """Gaussian Error Linear Unit.
    This is a smoother version of the RELU.
    Original paper: https://arxiv.org/abs/1606.08415
    Args:
      x: float Tensor to perform activation.
    Returns:
      `x` with the GELU activation applied.
    """
    cdf = 0.5 * (1.0 + tf.tanh(
        (np.sqrt(2 / np.pi) * (x + 0.044715 * tf.pow(x, 3)))))
    return x * cdf

def shape_list(x):
    """Return list of dims, statically where possible."""
    x = tf.convert_to_tensor(x)
    
    # If unknown rank, return dynamic shape
    if x.get_shape().dims is None:
        return tf.shape(x)
    
    static = x.get_shape().as_list()
    shape = tf.shape(x)
    
    ret = []
    for i in range(len(static)):
        dim = static[i]
        if dim is None:
            dim = shape[i]
        ret.append(dim)
    return ret

def get_timing_signal_1d(length,
                     channels,
                     min_timescale=1.0,
                     max_timescale=1.0e4,
                     start_index=0):
    """Gets a bunch of sinusoids of different frequencies.
    Each channel of the input Tensor is incremented by a sinusoid of a different
    frequency and phase.
    This allows attention to learn to use absolute and relative positions.
    Timing signals should be added to some precursors of both the query and the
    memory inputs to attention.
    The use of relative position is possible because sin(x+y) and cos(x+y) can be
    expressed in terms of y, sin(x) and cos(x).
    In particular, we use a geometric sequence of timescales starting with
    min_timescale and ending with max_timescale.  The number of different
    timescales is equal to channels / 2. For each timescale, we
    generate the two sinusoidal signals sin(timestep/timescale) and
    cos(timestep/timescale).  All of these sinusoids are concatenated in
    the channels dimension.
    Args:
    length: scalar, length of timing signal sequence.
    channels: scalar, size of timing embeddings to create. The number of
        different timescales is equal to channels / 2.
    min_timescale: a float
    max_timescale: a float
    start_index: index of first position
    Returns:
    a Tensor of timing signals [1, length, channels]
    """
    position = tf.to_float(tf.range(length) + start_index)
    num_timescales = channels // 2
    log_timescale_increment = (
      math.log(float(max_timescale) / float(min_timescale)) /
      (tf.to_float(num_timescales) - 1))
    inv_timescales = min_timescale * tf.exp(
      tf.to_float(tf.range(num_timescales)) * -log_timescale_increment)
    scaled_time = tf.expand_dims(position, 1) * tf.expand_dims(inv_timescales, 0)
    signal = tf.concat([tf.sin(scaled_time), tf.cos(scaled_time)], axis=1)
    signal = tf.pad(signal, [[0, 0], [0, tf.mod(channels, 2)]])
    signal = tf.reshape(signal, [1, length, channels])
    return signal


def add_timing_signal_1d(x,
                     min_timescale=1.0,
                     max_timescale=1.0e4,
                     start_index=0):
    """Adds a bunch of sinusoids of different frequencies to a Tensor.
    Each channel of the input Tensor is incremented by a sinusoid of a different
    frequency and phase.
    This allows attention to learn to use absolute and relative positions.
    Timing signals should be added to some precursors of both the query and the
    memory inputs to attention.
    The use of relative position is possible because sin(x+y) and cos(x+y) can be
    experessed in terms of y, sin(x) and cos(x).
    In particular, we use a geometric sequence of timescales starting with
    min_timescale and ending with max_timescale.  The number of different
    timescales is equal to channels / 2. For each timescale, we
    generate the two sinusoidal signals sin(timestep/timescale) and
    cos(timestep/timescale).  All of these sinusoids are concatenated in
    the channels dimension.
    Args:
      x: a Tensor with shape [batch, length, channels]
      min_timescale: a float
      max_timescale: a float
      start_index: index of first position
    Returns:
      a Tensor the same shape as x.
    """
    batchs = shape_list(x)[0]
    length = shape_list(x)[1]
    channels = shape_list(x)[2]
    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale,
                                                   start_index)
    _signal = tf.tile(signal,[batchs,1,1])
    _concat = tf.concat([x,_signal],axis=2)
    # _concat = tf.quantized_concat(2,[x,signal],input_mins=[-10,-10],input_maxes=[10,10])[0]
    print("##",_concat)
    return  _concat


def get_activation(activation_string):
    """Maps a string to a Python function, e.g., "relu" => `tf.nn.relu`.
    Args:
      activation_string: String name of the activation function.
    Returns:
      A Python function corresponding to the activation function. If
      `activation_string` is None, empty, or "linear", this will return None.
      If `activation_string` is not a string, it will return `activation_string`.
    Raises:
      ValueError: The `activation_string` does not correspond to a known
        activation.
    """
    
    # We assume that anything that"s not a string is already an activation
    # function, so we just return it.
    if not isinstance(activation_string, six.string_types):
        return activation_string
    
    if not activation_string:
        return None
    
    act = activation_string.lower()
    if act == "linear":
        return None
    elif act == "relu":
        return tf.nn.relu
    elif act == "gelu":
        return gelu
    elif act == "tanh":
        return tf.tanh
    else:
        raise ValueError("Unsupported activation: %s" % act)
    
def dropout(input_tensor, dropout_prob):
    """Perform dropout.
    Args:
      input_tensor: float Tensor.
      dropout_prob: Python float. The probability of dropping out a value (NOT of
        *keeping* a dimension as in `tf.nn.dropout`).
    Returns:
      A version of `input_tensor` with dropout applied.
    """
    if dropout_prob is None or dropout_prob == 0.0:
        return input_tensor
    
    output = tf.nn.dropout(input_tensor, 1.0 - dropout_prob)
    return output

def layer_norm(input_tensor, name=None):
    """Run layer normalization on the last dimension of the tensor."""
    return tf.contrib.layers.layer_norm(
        inputs=input_tensor, begin_norm_axis=-1, begin_params_axis=-1, scope=name)


def layer_norm_and_dropout(input_tensor, dropout_prob, name=None):
    """Runs layer normalization followed by dropout."""
    output_tensor = layer_norm(input_tensor, name)
    output_tensor = dropout(output_tensor, dropout_prob)
    return output_tensor


def create_initializer(initializer_range=0.02):
    """Creates a `truncated_normal_initializer` with the given range."""
    return tf.truncated_normal_initializer(stddev=initializer_range)

def reshape_to_matrix(input_tensor):
    """Reshapes a >= rank 2 tensor to a rank 2 tensor (i.e., a matrix)."""
    ndims = input_tensor.shape.ndims
    if ndims < 2:
        raise ValueError("Input tensor must have at least rank 2. Shape = %s" %
                       (input_tensor.shape))
    if ndims == 2:
        return input_tensor
    
    width = input_tensor.shape[-1]
    output_tensor = tf.reshape(input_tensor, [-1, width])
    return output_tensor


def reshape_from_matrix(output_tensor, orig_shape_list):
    """Reshapes a rank 2 tensor back to its original rank >= 2 tensor."""
    if len(orig_shape_list) == 2:
        return output_tensor
    
    output_shape = shape_list(output_tensor)
    
    orig_dims = orig_shape_list[0:-1]
    width = output_shape[-1]
    
    return tf.reshape(output_tensor, orig_dims + [width])

def assert_rank(tensor, expected_rank, name=None):
    """Raises an exception if the tensor rank is not of the expected rank.
    Args:
      tensor: A tf.Tensor to check the rank of.
      expected_rank: Python integer or list of integers, expected rank.
      name: Optional name of the tensor for the error message.
    Raises:
      ValueError: If the expected shape doesn't match the actual shape.
    """
    if name is None:
        name = tensor.name
    
    expected_rank_dict = {}
    if isinstance(expected_rank, six.integer_types):
        expected_rank_dict[expected_rank] = True
    else:
        for x in expected_rank:
            expected_rank_dict[x] = True
    
    actual_rank = tensor.shape.ndims
    if actual_rank not in expected_rank_dict:
        scope_name = tf.get_variable_scope().name
        raise ValueError(
          "For the tensor `%s` in scope `%s`, the actual rank "
          "`%d` (shape = %s) is not equal to the expected rank `%s`" %
          (name, scope_name, actual_rank, str(tensor.shape), str(expected_rank)))

def get_shape_list(tensor, expected_rank=None, name=None):
    """Returns a list of the shape of tensor, preferring static dimensions.
    Args:
      tensor: A tf.Tensor object to find the shape of.
      expected_rank: (optional) int. The expected rank of `tensor`. If this is
        specified and the `tensor` has a different rank, and exception will be
        thrown.
      name: Optional name of the tensor for the error message.
    Returns:
      A list of dimensions of the shape of tensor. All static dimensions will
      be returned as python integers, and dynamic dimensions will be returned
      as tf.Tensor scalars.
    """
    if name is None:
        name = tensor.name
    
    if expected_rank is not None:
        assert_rank(tensor, expected_rank, name)
    
    shape = tensor.shape.as_list()
    
    non_static_indexes = []
    for (index, dim) in enumerate(shape):
        if dim is None:
            non_static_indexes.append(index)
    
    if not non_static_indexes:
        return shape
    
    dyn_shape = tf.shape(tensor)
    for index in non_static_indexes:
        shape[index] = dyn_shape[index]
    return shape

def attention_layer(from_tensor,
                    to_tensor,
                    attention_mask=None,
                    num_attention_heads=1,
                    size_per_head=10,
                    query_act=None,
                    key_act=None,
                    value_act=None,
                    attention_probs_dropout_prob=0.0,
                    initializer_range=0.02,
                    do_return_2d_tensor=False,
                    batch_size=None,
                    from_seq_length=None,
                    to_seq_length=None):
  """Performs multi-headed attention from `from_tensor` to `to_tensor`.
  This is an implementation of multi-headed attention based on "Attention
  is all you Need". If `from_tensor` and `to_tensor` are the same, then
  this is self-attention. Each timestep in `from_tensor` attends to the
  corresponding sequence in `to_tensor`, and returns a fixed-with vector.
  This function first projects `from_tensor` into a "query" tensor and
  `to_tensor` into "key" and "value" tensors. These are (effectively) a list
  of tensors of length `num_attention_heads`, where each tensor is of shape
  [batch_size, seq_length, size_per_head].
  Then, the query and key tensors are dot-producted and scaled. These are
  softmaxed to obtain attention probabilities. The value tensors are then
  interpolated by these probabilities, then concatenated back to a single
  tensor and returned.
  In practice, the multi-headed attention are done with transposes and
  reshapes rather than actual separate tensors.
  Args:
    from_tensor: float Tensor of shape [batch_size, from_seq_length,
      from_width].
    to_tensor: float Tensor of shape [batch_size, to_seq_length, to_width].
    attention_mask: (optional) int32 Tensor of shape [batch_size,
      from_seq_length, to_seq_length]. The values should be 1 or 0. The
      attention scores will effectively be set to -infinity for any positions in
      the mask that are 0, and will be unchanged for positions that are 1.
    num_attention_heads: int. Number of attention heads.
    size_per_head: int. Size of each attention head.
    query_act: (optional) Activation function for the query transform.
    key_act: (optional) Activation function for the key transform.
    value_act: (optional) Activation function for the value transform.
    attention_probs_dropout_prob: (optional) float. Dropout probability of the
      attention probabilities.
    initializer_range: float. Range of the weight initializer.
    do_return_2d_tensor: bool. If True, the output will be of shape [batch_size
      * from_seq_length, num_attention_heads * size_per_head]. If False, the
      output will be of shape [batch_size, from_seq_length, num_attention_heads
      * size_per_head].
    batch_size: (Optional) int. If the input is 2D, this might be the batch size
      of the 3D version of the `from_tensor` and `to_tensor`.
    from_seq_length: (Optional) If the input is 2D, this might be the seq length
      of the 3D version of the `from_tensor`.
    to_seq_length: (Optional) If the input is 2D, this might be the seq length
      of the 3D version of the `to_tensor`.
  Returns:
    float Tensor of shape [batch_size, from_seq_length,
      num_attention_heads * size_per_head]. (If `do_return_2d_tensor` is
      true, this will be of shape [batch_size * from_seq_length,
      num_attention_heads * size_per_head]).
  Raises:
    ValueError: Any of the arguments or tensor shapes are invalid.
  """
  
  def transpose_for_scores(input_tensor, batch_size, num_attention_heads,
                           seq_length, width):
    output_tensor = tf.reshape(
        input_tensor, [batch_size, seq_length, num_attention_heads, width])
  
    output_tensor = tf.transpose(output_tensor, [0, 2, 1, 3])
    return output_tensor
  
  from_shape = get_shape_list(from_tensor, expected_rank=[2, 3])
  to_shape = get_shape_list(to_tensor, expected_rank=[2, 3])
  
  if len(from_shape) != len(to_shape):
    raise ValueError(
        "The rank of `from_tensor` must match the rank of `to_tensor`.")
  
  if len(from_shape) == 3:
    batch_size = from_shape[0]
    from_seq_length = from_shape[1]
    to_seq_length = to_shape[1]
  elif len(from_shape) == 2:
    if (batch_size is None or from_seq_length is None or to_seq_length is None):
      raise ValueError(
          "When passing in rank 2 tensors to attention_layer, the values "
          "for `batch_size`, `from_seq_length`, and `to_seq_length` "
          "must all be specified.")
  
  # Scalar dimensions referenced here:
  #   B = batch size (number of sequences)
  #   F = `from_tensor` sequence length
  #   T = `to_tensor` sequence length
  #   N = `num_attention_heads`
  #   H = `size_per_head`
  
  from_tensor_2d = reshape_to_matrix(from_tensor)
  to_tensor_2d = reshape_to_matrix(to_tensor)
  
  # `query_layer` = [B*F, N*H]
  '''
  query_matrix = tf.get_variable(name="query",shape=(shape_list(from_tensor_2d)[-1],num_attention_heads * size_per_head),initializer=create_initializer(initializer_range))
  query_layer = tf.matmul(from_tensor_2d,query_matrix)
  if query_act is not None:
      query_layer = query_act(query_layer)
      
  key_matrix = tf.get_variable(name="key",shape=(shape_list(from_tensor_2d)[-1],num_attention_heads * size_per_head),initializer=create_initializer(initializer_range))
  key_layer = tf.matmul(from_tensor_2d,key_matrix)
  if key_act is not None:
      key_layer =key_act(key_layer)
      
  value_matrix = tf.get_variable(name="value",shape=(shape_list(from_tensor_2d)[-1],num_attention_heads * size_per_head),initializer=create_initializer(initializer_range))
  value_layer = tf.matmul(from_tensor_2d,value_matrix)
  if value_act is not None:
      value_layer = value_act(value_layer)
      
  '''
  query_layer = tf.layers.dense(
      from_tensor_2d,
      num_attention_heads * size_per_head,
      activation=query_act,
      name="query",
      kernel_initializer=create_initializer(initializer_range))
  
  # `key_layer` = [B*T, N*H]
  key_layer = tf.layers.dense(
      to_tensor_2d,
      num_attention_heads * size_per_head,
      activation=key_act,
      name="key",
      kernel_initializer=create_initializer(initializer_range))
  
  # `value_layer` = [B*T, N*H]
  value_layer = tf.layers.dense(
      to_tensor_2d,
      num_attention_heads * size_per_head,
      activation=value_act,
      name="value",
      kernel_initializer=create_initializer(initializer_range))
  
  # `query_layer` = [B, N, F, H]
  query_layer = transpose_for_scores(query_layer, batch_size,
                                     num_attention_heads, from_seq_length,
                                     size_per_head)
  
  # `key_layer` = [B, N, T, H]
  key_layer = transpose_for_scores(key_layer, batch_size, num_attention_heads,
                                   to_seq_length, size_per_head)
  
  # Take the dot product between "query" and "key" to get the raw
  # attention scores.
  # `attention_scores` = [B, N, F, T]
  attention_scores = tf.matmul(query_layer, key_layer, transpose_b=True)
  attention_scores = tf.multiply(attention_scores,
                                 1.0 / math.sqrt(float(size_per_head)))
  print(attention_scores)
  if attention_mask is not None:
      # `attention_mask` = [B, 1, F, T]
      attention_mask = tf.expand_dims(attention_mask, axis=[1])
      
      # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
      # masked positions, this operation will create a tensor which is 0.0 for
      # positions we want to attend and -10000.0 for masked positions.
      adder = (1.0 - tf.cast(attention_mask, tf.float32)) * -10000.0
      
      # Since we are adding it to the raw scores before the softmax, this is
      # effectively the same as removing these entirely.
      attention_scores += adder
  
  # Normalize the attention scores to probabilities.
  # `attention_probs` = [B, N, F, T]
  #   B = batch size (number of sequences)
  #   F = `from_tensor` sequence length
  #   T = `to_tensor` sequence length
  #   N = `num_attention_heads`
  #   H = `size_per_head`
  #attention_scores = tf.reshape(attention_scores,[batch_size,num_attention_heads,from_seq_length,to_seq_length])  
  attention_probs = tf.nn.softmax(attention_scores)
  
  # This is actually dropping out entire tokens to attend to, which might
  # seem a bit unusual, but is taken from the original Transformer paper.
  attention_probs = dropout(attention_probs, attention_probs_dropout_prob)
  
  # `value_layer` = [B, T, N, H]
  value_layer = tf.reshape(
      value_layer,
      [batch_size, to_seq_length, num_attention_heads, size_per_head])
  
  # `value_layer` = [B, N, T, H]
  value_layer = tf.transpose(value_layer, [0, 2, 1, 3])
  
  # `context_layer` = [B, N, F, H]
  context_layer = tf.matmul(attention_probs, value_layer)
  
  # `context_layer` = [B, F, N, H]
  context_layer = tf.transpose(context_layer, [0, 2, 1, 3])
  
  if do_return_2d_tensor:
    # `context_layer` = [B*F, N*H]
    context_layer = tf.reshape(
        context_layer,
        [batch_size * from_seq_length, num_attention_heads * size_per_head])
  else:
    # `context_layer` = [B, F, N*H]
    context_layer = tf.reshape(
        context_layer,
        [batch_size, from_seq_length, num_attention_heads * size_per_head])
  
  return context_layer
        
def transformer_model(input_tensor,
                      attention_mask=None,
                      hidden_size=120,
                      num_hidden_layers=1,
                      num_attention_heads=6,
                      intermediate_size=256,
                      intermediate_act_fn=gelu,
                      hidden_dropout_prob=0.1,
                      attention_probs_dropout_prob=0.1,
                      initializer_range=0.02,
                      do_return_all_layers=False,
                      name=None):
    
  input_tensor = add_timing_signal_1d(input_tensor,max_timescale=1000)
  
  if hidden_size % num_attention_heads != 0:
      raise ValueError(
          "The hidden size (%d) is not a multiple of the number of attention "
          "heads (%d)" % (hidden_size, num_attention_heads))
  
  attention_head_size = int(hidden_size / num_attention_heads)
  input_shape = get_shape_list(input_tensor, expected_rank=3)
  batch_size = input_shape[0]
  seq_length = input_shape[1]
  input_width = input_shape[2]
  
  # The Transformer performs sum residuals on all layers so the input needs
  # to be the same as the hidden size.
  if input_width != hidden_size:
    raise ValueError("The width of the input tensor (%d) != hidden size (%d)" %
                     (input_width, hidden_size))
  
  # We keep the representation as a 2D tensor to avoid re-shaping it back and
  # forth from a 3D tensor to a 2D tensor. Re-shapes are normally free on
  # the GPU/CPU but may not be free on the TPU, so we want to minimize them to
  # help the optimizer.
  prev_output = reshape_to_matrix(input_tensor)
  
  all_layer_outputs = []
  if name is not None:
    _name = str(name)+"encoder"
  else:
    _name = "encoder"
  with tf.variable_scope(_name,reuse=tf.AUTO_REUSE):
    for layer_idx in range(num_hidden_layers):
      with tf.variable_scope("layer_%d" % layer_idx,reuse=tf.AUTO_REUSE):
        layer_input = prev_output
    
        with tf.variable_scope("attention",reuse=tf.AUTO_REUSE):
          attention_heads = []
          with tf.variable_scope("self",reuse=tf.AUTO_REUSE):
            attention_head = attention_layer(
                from_tensor=layer_input,
                to_tensor=layer_input,
                attention_mask=attention_mask,
                num_attention_heads=num_attention_heads,
                size_per_head=attention_head_size,
                attention_probs_dropout_prob=attention_probs_dropout_prob,
                initializer_range=initializer_range,
                do_return_2d_tensor=True,
                batch_size=batch_size,
                from_seq_length=seq_length,
                to_seq_length=seq_length)
            attention_heads.append(attention_head)
    
          attention_output = None
          if len(attention_heads) == 1:
            attention_output = attention_heads[0]
          else:
            # In the case where we have other sequences, we just concatenate
            # them to the self-attention head before the projection.
            attention_output = tf.concat(attention_heads, axis=-1)
    
          # Run a linear projection of `hidden_size` then add a residual
          # with `layer_input`.
          with tf.variable_scope("output",reuse=tf.AUTO_REUSE):
            attention_output = tf.layers.dense(
                attention_output,
                hidden_size,
                kernel_initializer=create_initializer(initializer_range))
            attention_output = dropout(attention_output, hidden_dropout_prob)
            attention_output = layer_norm(attention_output + layer_input)
    
        # The activation is only applied to the "intermediate" hidden layer.
        with tf.variable_scope("intermediate",reuse=tf.AUTO_REUSE):
          intermediate_output = tf.layers.dense(
              attention_output,
              intermediate_size,
              activation=intermediate_act_fn,
              kernel_initializer=create_initializer(initializer_range))
    
        # Down-project back to `hidden_size` then add the residual.
        with tf.variable_scope("output",reuse=tf.AUTO_REUSE):
          layer_output = tf.layers.dense(
              intermediate_output,
              hidden_size,
              kernel_initializer=create_initializer(initializer_range))
          layer_output = dropout(layer_output, hidden_dropout_prob)
          layer_output = layer_norm(layer_output + attention_output)
          prev_output = layer_output
          all_layer_outputs.append(layer_output)
    
    if do_return_all_layers:
      final_outputs = []
      for layer_output in all_layer_outputs:
        final_output = reshape_from_matrix(layer_output, input_shape)
        final_outputs.append(final_output)
      return final_outputs
    else:
      final_output = reshape_from_matrix(prev_output, input_shape)
      return final_output
                    
    
    
        
# def getBiLSTMModel(input_shape,vocab,embedding_weights,classes,use_am=False):
#
#     assert len(input_shape)==3
#     list_input = []
#     for i in range(input_shape[0]):
#         list_input.append(layers.Input(shape=(input_shape[1],),dtype=tf.int32))
#     print(list_input)
#     list_embedding = []
#
#     embedding = layers.Embedding(len(vocab),input_shape[2],weights=[embedding_weights] if embedding_weights is not None else None,trainable=True,name="char_embeding")
#     for i in range(len(list_input)):
#         list_embedding.append(embedding(list_input[i]))
#     #
#     bert_layer = layers.Lambda(transformer_model,trainable=True,name="bert")
#     set_variables = set()
#     for v in tf.trainable_variables():
#         set_variables.add(v.name)
#     list_bert = []
#     for i in range(len(list_embedding)):
#         list_bert.append(bert_layer(list_embedding[i]))
#     #set bert_weights to trainable
#     bert_weights = []
#     for v in tf.trainable_variables():
#         if v.name not in set_variables:
#             bert_weights.append(v)
#     bert_layer._trainable_weights = bert_weights
#
#     list_w2v = list_bert
#     list_lstm = []
#
#     if use_am:
#         for i in range(len(list_w2v)):
#             list_lstm.append(Attention()(layers.Bidirectional(layers.LSTM(120,activation="relu",return_sequences=True))(list_w2v[i])))
#     else:
#         for i in range(len(list_w2v)):
#             list_lstm.append(layers.Bidirectional(layers.LSTM(24,activation="relu"))(list_w2v[i]))
#
#
#     list_matrix = []
#     for i in range(len(list_lstm)):
#         list_matrix.append(layers.Dense(classes*2,activation="relu")(list_lstm[i]))
#
#     if len(list_matrix)>1:
#         ave = layers.merge(list_matrix,mode="concat")
#
#         dropout = layers.Dropout(0.4)(ave)
#     else:
#         dropout = layers.Dropout(0.4)(list_matrix[0])
#
#
#     matrix = layers.Dense(classes*10,activation="sigmoid")(dropout)
#
#     out = layers.Dense(classes,activation="softmax")(matrix)
#
#     model = models.Model(list_input,out)
#
#     model.compile(optimizer=optimizers.Adam(lr=0.00002),loss=losses.categorical_crossentropy,metrics=[precision,recall,f1_score])
#
#     model.summary()
#
#     return model

def getBiLSTMModel(input_shape,vocab,embedding_weights,classes,use_am=False):
    def resize(x):
        _shape = shape_list(x)
        print("#$",_shape)
        x1 = tf.reshape(x,[_shape[0],3,_shape[1]//3,_shape[2]])  # type: object
        print("--")
        x2 = tf.transpose(x1,[1,0,2,3])
        x_l,x_c,x_r = tf.split(x2,[1,1,1],axis=0)
        return [tf.squeeze(x_l,axis=0),tf.squeeze(x_c,axis=0),tf.squeeze(x_r,axis=0)]

    def resize_input(x):
        _shape = shape_list(x)
        x1 = tf.reshape(x,[_shape[0],3,_shape[1]//3])  # type: object
        x2 = tf.transpose(x1,[1,0,2])
        x_l,x_c,x_r = tf.split(x2,[1,1,1],axis=0)
        return [tf.squeeze(x_l,axis=0),tf.squeeze(x_c,axis=0),tf.squeeze(x_r,axis=0)]

    # assert len(input_shape)==3
    list_input = []
    for i in range(input_shape[0]):
        list_input.append(layers.Input(shape=(input_shape[1],),dtype=tf.int32,name="input%d"%(i)))
    print("list_input",list_input)
    list_embedding = []

    # if len(list_input)==1:
    #     list_resizeinput = layers.Lambda(resize_input)(list_input[0])
    # else:

    # concat_input = layers.Lambda(lambda x:tf.concat(x,axis=-1))(list_input)
    # embedding_input = [concat_input]
    embedding_input = list_input

    embedding = layers.Embedding(len(vocab),input_shape[2],weights=[embedding_weights] if embedding_weights is not None else None,trainable=True,name="char_embeding")
    for i in range(len(embedding_input)):
        print(i)
        list_embedding.append(embedding(embedding_input[i]))
        print(list_embedding)


    #
    set_variables = set()
    for v in tf.trainable_variables():
        set_variables.add(v.name)

    # list_bert = []
    # for i in range(len(list_embedding)):
    #     for v in tf.trainable_variables():
    #         set_variables.add(v.name)
    #     bert_layer = layers.Lambda(lambda x:transformer_model(input_tensor=x,name="bert%d"%(i)),trainable=True,name="bert%d"%(i))
    #     list_bert.append(bert_layer(list_embedding[i]))
    #     #set bert_weights to trainable
    #     bert_weights = []
    #     for v in tf.trainable_variables():
    #         if v.name not in set_variables:
    #             print("++++",v.name)
    #             bert_weights.append(v)
    #     bert_layer._trainable_weights = bert_weights


    bert_layer = layers.Lambda(lambda x:transformer_model(input_tensor=x,name="bert%d"%(i)),trainable=True,name="bert%d"%(0))
    list_bert = []
    for i in range(len(list_embedding)):
        list_bert.append(bert_layer(list_embedding[i]))
        #set bert_weights to trainable
    bert_weights = []
    for v in tf.trainable_variables():
        if v.name not in set_variables:
            print("++++",v.name)
            bert_weights.append(v)
    bert_layer._trainable_weights = bert_weights

    print("##",list_bert)


    # context_embedding = []
    # list_kernel = [5,8]
    # for i in range(len(list_bert)):
    #     list_temp = []
    #     for kernel in list_kernel:
    #         list_temp.append(layers.Conv1D(3,kernel,strides=1,padding="same",activation="relu")(list_bert[i]))
    #     context_embedding.append(layers.Dense(12,activation="relu")(layers.Flatten()(layers.merge(list_temp,mode="concat"))))

    # context_embedding = [layers.GlobalMaxPool1D()(item) for item in list_bert]



    # _resize = layers.Lambda(lambda x:resize(x))(list_bert[0])
    list_w2v = list_bert
    list_lstm = []

    if use_am:
        for i in range(len(list_w2v)):
            list_lstm.append(Attention()(layers.Bidirectional(layers.LSTM(120,activation="relu",return_sequences=True))(list_w2v[i])))
    else:
        for i in range(len(list_w2v)):
            list_lstm.append(layers.Bidirectional(layers.LSTM(24,activation="relu"))(list_w2v[i]))

    # list_avg = []
    # for i in range(len(list_lstm)):
    #     list_avg.append(layers.GlobalAveragePooling1D()(list_lstm[i]))


    list_matrix = []
    for i in range(len(list_lstm)):
        list_matrix.append(layers.Dense(12,activation="relu")(list_lstm[i]))
    # list_matrix.extend(context_embedding)

    if len(list_matrix)>1:
        ave = layers.merge(list_matrix,mode="concat")

        dropout = layers.Dropout(0.2)(ave)
    else:
        dropout = layers.Dropout(0.2)(list_matrix[0])


    matrix = layers.Dense(48,activation="tanh")(dropout)
    out = layers.Dense(classes,activation="softmax")(matrix)

    # out = layers.Dense(classes,activation="sigmoid")(dropout)
    # out = layers.Lambda(lambda x:layers.activations.softmax(x))(out)

    model = models.Model(list_input,out)

    model.compile(optimizer=optimizers.Adam(lr=0.01),loss=losses.categorical_crossentropy,metrics=[precision,recall,f1_score])

    model.summary()

    return model

def getBiLSTMModel_entity(input_shape,vocab,embedding_weights,classes):

    list_input = []
    for i in range(input_shape[0]):
        list_input.append(layers.Input(shape=(input_shape[1],),dtype=tf.int32,name="input%d"%(i)))
    print("list_input",list_input)
    list_embedding = []

    embedding_input = list_input

    embedding = layers.Embedding(len(vocab),input_shape[2],weights=[embedding_weights] if embedding_weights is not None else None,trainable=True,name="char_embeding")
    for i in range(len(embedding_input)):
        print(i)
        list_embedding.append(embedding(embedding_input[i]))
        print(list_embedding)


    list_w2v = list_embedding
    list_lstm = []

    for i in range(len(list_w2v)):
        list_lstm.append(layers.Bidirectional(layers.LSTM(24,activation="relu"))(list_w2v[i]))

    # list_avg = []
    # for i in range(len(list_lstm)):
    #     list_avg.append(layers.GlobalAveragePooling1D()(list_lstm[i]))


    list_matrix = []
    for i in range(len(list_lstm)):
        list_matrix.append(layers.Dense(12,activation="relu")(list_lstm[i]))

    if len(list_matrix)>1:
        ave = layers.merge(list_matrix,mode="concat")

        dropout = layers.Dropout(0.2)(ave)
    else:
        dropout = layers.Dropout(0.2)(list_matrix[0])


    matrix = layers.Dense(classes*10,activation="relu")(dropout)
    out = layers.Dense(classes,activation="softmax")(matrix)

    # out = layers.Dense(classes,activation="sigmoid")(dropout)
    # out = layers.Lambda(lambda x:layers.activations.softmax(x))(out)

    model = models.Model(list_input,out)

    model.compile(optimizer=optimizers.Adam(lr=0.00001),loss=losses.categorical_crossentropy,metrics=[precision,recall,f1_score])

    model.summary()

    return model





if __name__=="__main__":
  #getTextCNNModel((3,100,60),[1,2,3,4,5],None,2)
  model = getBiLSTMModel((3,100,256),fool_char_to_id.keys(),None,3,use_am=False)
  #getBiLSTMModel_entity((20,20,3,100,60),[1,2,3,4,5],None,6)