cnn.py

x = Input(shape=(max_len,))
emb = Embedding(max_features,
                embedding_dims,
                input_length=max_len))(x)
                
for filter_len in [2, 3, 4]:
    branch = Sequential()
    branch.add(Convolution1D(nb_filter=filters,
                             filter_length=filter_len,
                             border_mode='valid',
                             activation='relu',
                             subsample_length=subsample_length,
                             input_shape=(max_len, embedding_dims))
    branch.add(MaxPooling1D(pool_length=pooling_length))
    branch.add(Flatten())
    branches.append(branch(emb))
merged = Merge(branches, mode='concat')
y = Dense(1, activation='sigmoid')(merged)
model = Model(inputs=x, outputs=y)