Building the Autoencoder

Define the encoder stacked_encoder with a flattened input layer, along with 2 dense layers - one with 100 neurons - selu activation while the other with 30 neurons - selu activation function. We shall add these layers to keras.models.Sequential.

stacked_encoder = keras.models.Sequential([
    keras.layers.Flatten(input_shape=[28, 28]),
    keras.layers.Dense(100, activation="selu"),
    keras.layers.Dense(30, activation="selu"),
])

Similarly, we shall define the encoder stacked_decoder with 2 dense layers - one with 100 neurons - selu activation while the other with 28*28 neurons - sigmoid activation function, followed by an output layer of shape 28 x 28. We shall add these layers to keras.models.Sequential.

stacked_decoder = keras.models.Sequential([
    keras.layers.Dense(100, activation="selu", input_shape=[30]),
    keras.layers.Dense(28 * 28, activation="sigmoid"),
    keras.layers.Reshape([28, 28])
])

We shall now club the stacked_encoder and stacked_decoder using keras.models.Sequential to form our complete autoencoder stacked_ae.

stacked_ae = keras.models.Sequential([stacked_encoder, stacked_decoder])

Now, we compile the stacked_ae by using compile. We shall set "binary_crossentropy" loss, keras.optimizers.SGD(lr=1.5) optimizer and rounded_accuracy(which we have previously defined) metric.

stacked_ae.<< your code comes here >>(loss="binary_crossentropy",
           optimizer=<< your code comes here >>, metrics=[rounded_accuracy])