Function bcnn

Attention_Based_CNN (ABCNN)/model_abcnn.py:144–223

(embedding_matrix, dimx=50, dimy=50, nb_filter = 120, embedding_dim = 50, 
                      filter_length = (50,4), depth = 1, shared = 0,
                      opt_params = [0.0008,'adam'])

def bcnn(embedding_matrix, dimx=50, dimy=50, nb_filter = 120, embedding_dim = 50, 
                      filter_length = (50,4), depth = 1, shared = 0,
                      opt_params = [0.0008,'adam']):

#if True:
    print 'Model Uses BCNN......'
    
    inpx = Input(shape=(dimx,),dtype='int32',name='inpx')   
    inpy = Input(shape=(dimy,),dtype='int32',name='inpy')
    
    x = word2vec_embedding_layer(embedding_matrix,train=False)(inpx)
    y = word2vec_embedding_layer(embedding_matrix,train=False)(inpy)
    
    x = Permute((2,1))(x)
    y = Permute((2,1))(y)

    conv1 = Reshape((embedding_dim,dimx,1))(x)
    conv2 = Reshape((embedding_dim,dimy,1))(y)   
       
    channel_1, channel_2 = [], []
    
    for dep in range(depth):
        
        #filter_width = filter_widths[dep]
        
        #conv1 = ZeroPadding2D((filter_width - 1, 0))(conv1)
        #conv2 = ZeroPadding2D((filter_width - 1, 0))(conv2)
        
        if shared:
            conv = Conv2D(nb_filter=nb_filter, kernel_size = filter_length, activation='relu',
                        data_format = 'channels_last',border_mode="valid")
            ques = conv(conv1)
            ans = conv(conv2)
                
        else:
            ques = Conv2D(nb_filter=nb_filter, kernel_size = filter_length, activation='relu',
                    data_format = 'channels_last',border_mode="valid")(conv1)
            ans = Conv2D(nb_filter, kernel_size = filter_length, activation='relu',
                    data_format="channels_last",border_mode="valid")(conv2)
                    
            
        ques = Dropout(0.5)(ques)
        ans = Dropout(0.5)(ans)
        channel_1.append(GlobalMaxPooling2D()(ques))
        channel_2.append(GlobalMaxPooling2D()(ans))
        
        #channel_1.append(Reshape((ques._keras_shape[2]*ans._keras_shape[3]))(AveragePooling2D(4))(ques))
        #channel_2.appendFlatten()((AveragePooling2D(4))(ans))

       
    #reg1 = reg2 = 0.00002
    
    h1 = channel_1.pop(-1)
    if channel_1:
        h1 = merge([h1] + channel_1, mode="concat")

    h2 = channel_2.pop(-1)
    if channel_2: