Encoder.py

import matplotlib.pyplot as plt
from keras.layers import Input, Dense
from keras.models import model_from_json
from keras.models import Model
from keras.utils.layer_utils import layer_from_config
from Autoencoder import Autoencoder


class Encoder(Autoencoder):
    def __init__(self, path_to_model, path_to_weights):
        self.autoencoder = model_from_json(open(path_to_model).read())
        self.autoencoder.load_weights(path_to_weights)
        self.autoencoder.summary()
        self.input_img = Input(shape=(1240,))
        self.encoded = Dense(
            self.autoencoder.layers[1].output_dim,
            activation=self.autoencoder.layers[1].activation,
            weights=self.autoencoder.layers[1].get_weights()
        )(self.input_img)
        self.encoder = Model(input=self.input_img, output=self.encoded)
        self.encoder.summary()
        self.out_dim = self.encoder.layers[1].output_dim
        self.in_dim = 1240

    def encode(self, img):
        return self.encoder.predict(img)

    def draw(self, img):
        return self.autoencoder.predict(img)

    def decode(self, img):
        return self.decoder.predict(img)

    def fit(self, data, n_epochs=1000, val_split=0.1, batch_s=512):
        if use_pre_trained_model:
            print 'Fit is not necessary!'
            return
        autoencoder.fit(
            data,
            data,
            nb_epoch=n_epochs,
            batch_size=batch_s,
            shuffle=True,
            validation_split=val_split
        )