model.py

import tensorflow as tf
from tensorflow.python import keras

from keras.models import Model
from keras.optimizers import Adam, SGD
from keras.layers import Input, Dense, Flatten, Concatenate, Multiply, Dropout, Subtract, Embedding, LSTM, Lambda, BatchNormalization, Bidirectional, concatenate

from eval import Eval

class SiameseModel:
    def __call__(self, shape, vocab_size, max_len, embedding_dim, embedding_matrix):
        input_1 = Input(shape=shape)
        input_2 = Input(shape=shape)
        
        word_embedding_1 = Embedding(input_dim=vocab_size, weights=[embedding_matrix], output_dim=embedding_dim, input_length=max_len, trainable=False)(input_1)
        word_embedding_2 = Embedding(input_dim=vocab_size, weights=[embedding_matrix], output_dim=embedding_dim, input_length=max_len, trainable=False)(input_2)
        
        lstm_1 = Bidirectional(LSTM(128, return_sequences=True))(word_embedding_1)
        lstm_1 = Dropout(0.25)(lstm_1)
        lstm_1 = Bidirectional(LSTM(128, return_sequences=True))(lstm_1)
        lstm_1 = Dropout(0.25)(lstm_1)
        
        lstm_2 = Bidirectional(LSTM(128, return_sequences=True))(word_embedding_2)
        lstm_2 = Dropout(0.25)(lstm_2)
        lstm_2 = Bidirectional(LSTM(128, return_sequences=True))(lstm_2)
        lstm_2 = Dropout(0.25)(lstm_2)
        
        concat = concatenate([lstm_1, lstm_2])
        
        merged = BatchNormalization()(concat)
        merged = Dropout(0.25)(merged)
        
        merged = Dense(64, activation="relu")(merged)
        merged = BatchNormalization()(merged)
        merged = Dropout(0.25)(merged)
        
        output = Dense(1, activation="sigmoid")(merged)
        
        model = Model(inputs=[input_1, input_2], outputs=output)
        
        model.compile(loss="binary_crossentropy", 
                      metrics=["accuracy"], 
                      optimizer=Adam(0.0001)
                )
        
        print(model.summary())
        
        return model