import keras
from keras.layers import Input, Conv2D, MaxPool2D, Dense, Flatten, Dropout, Reshape, UpSampling2D
from keras.models import Model
from keras.datasets import mnist
from sklearn.datasets import fetch_olivetti_faces
from tensorflow.keras.utils import to_categorical
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.model_selection import train_test_split

(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = np.reshape(x_train, (len(x_train), 28, 28, 1))
x_test  = np.reshape(x_test,  (len(x_test),  28, 28, 1))

x_train / 255.0
x_test / 255.0

y_train = np.reshape(y_train, (len(y_train), 1, 1, 1))
y_test  = np.reshape(y_test,  (len(y_test), 1, 1, 1))

y_train.shape

input_layer = Input(shape=(28, 28, 1))
conv1 = Conv2D(256, (1,1), activation='relu')(input_layer)
pool1 = MaxPool2D()(conv1)
conv2 = Conv2D(128, (1,1), activation='elu')(pool1)
pool2 = MaxPool2D()(conv2)
encoded = Dense(1,activation='softmax')(pool2)

input_layer2 = Input(shape=(28, 28, 1))
x = Conv2D(128, (1,1), activation='relu')(input_layer2)
x = MaxPool2D()(x)
x = Conv2D(256, (1,1),activation='elu')(x)
x = MaxPool2D()(x)
decoded = Dense(1, activation='softmax')(x)

encoder = Model(input_layer, encoded)
decoder = Model(input_layer2, decoded)
autoencoder = Model(input_layer, decoder(encoder(input_layer)))

autoencoder.compile(optimizer='adam', loss='binary_crossentropy')

autoencoder.fit(x_train, y_train, epochs=2,
          batch_size=64, validation_data=[x_test, y_test])