BIDI_ML_INFO_EXTRACTION/BiddingKG/dl/form/model.py

'''
Created on 2019年3月26日

@author: User
'''

import os
import sys
sys.path.append(os.path.abspath("../.."))

from BiddingKG.dl.common.Utils import *
from keras import layers
from keras import models
from keras import losses,optimizers
import tensorflow as tf

def getBiLSTMModel():
    input = layers.Input(shape=(100,))
    
    lstm = layers.Bidirectional(layers.LSTM(128,return_sequences=False))(input)
    
    matrix = layers.Dense(48,activation="relu")(lstm)
    
    out = layers.Dense(2,activation="softmax")(matrix)
    
    model = models.Model(inputs=input,outputs=out)
    
    model.compile(optimizer=optimizers.Adam(lr=0.01), loss=losses.categorical_crossentropy,metrics=[precision,recall,f1_score])
    model.summary()
    return model

def getTextCNNModel():
    input = layers.Input(shape=(100,60))
    
    list_pool = []
    
    list_size = [10,15,20]

    for size in list_size:    
        c = layers.Conv1D(filters=4,kernel_size=size,strides=1,activation="relu")(input)
        p = layers.AvgPool1D(pool_size=int(c.shape[1]))(c)
        list_pool.append(p)
    
    concat = layers.merge(list_pool,mode="concat")
    
    flatten = layers.Flatten()(concat)
    
    matrix = layers.Dense(12,activation="relu")(flatten)
    
    
    out = layers.Dense(2,activation="softmax")(matrix)
    
    model = models.Model(inputs=input,outputs=out)
    model.compile(optimizer=optimizers.Adadelta(),loss=losses.categorical_crossentropy,metrics=[precision,recall,f1_score])
    model.summary()
    return model

def get_context_form_model(vocab_len,char_dim,lstm_dim,context_dim):
    input = tf.placeholder(shape=[None,9,30])

    with tf.variable_scope("embedding"):
        embedding = tf.get_variable("char_embedding",shape=[vocab_len,char_dim])
        embedding_input = tf.nn.embedding_lookup(embedding,input)

    with tf.variable_scope("bi_rnn"):
        resize_input = tf.reshape(embedding_input,[-1,30,char_dim])

        forward_cell = tf.contrib.rnn.BasicLSTMCell(lstm_dim, state_is_tuple=True)
        backward_cell = tf.contrib.rnn.BasicLSTMCell(lstm_dim,state_is_tuple=True)
        outputs,forward_output,backward_output = tf.nn.static_bidirectional_rnn(forward_cell,backward_cell,resize_input)
        bi_output = tf.concat([forward_output,backward_output],-1) #[-1,lstm_dim*2]


    with tf.variable_scope("context"):
        context_out = tf.reshape(bi_output,[-1,9,lstm_dim*2])
        context_v = tf.get_variable("context_v",shape=[lstm_dim*2,context_dim])

        context_emb = tf.nn.relu(tf.matmul(context_out,context_v))

        center_out = tf.slice(context_out,[0,4,0],[-1,1,-1])





if __name__=="__main__":
    #getBiLSTMModel()
    getTextCNNModel()