import numpy as np
import keras
from keras.models import Model
from keras.layers import Input, Dense

def get_model1():
    inputs = Input(shape=(20,))
    x = inputs
    x = Dense(15)(x)
    x = Dense(10)(x)
    x = Dense(30, activation='softmax')(x)
    model = Model(inputs=inputs, outputs=x)
    return model

def main():
    model1 = get_model1()
    model1.compile(loss='categorical_crossentropy', optimizer='sgd')
    print(model1.summary())
    # Assume a scenario where user sets weights of model1 via some
    # pretraining strategy, such as stacked DAEs, and the size of the last
    # layer is inconsequential, since users typically ignore it in pretraining
    # Now, after it is trained, we save the weights
    # (just using the initial weights here)
    model1.save_weights('pretrained_weights.h5')

if __name__ == '__main__':
    main()