25 lines
848 B
Python
25 lines
848 B
Python
from keras.layers import Input, Dense
|
|
from keras.models import Model
|
|
import keras.datasets.mnist as mnist
|
|
|
|
(xtrain, ytrain), (xtest, ytest) = mnist.load_data()
|
|
|
|
# this is the size of our encoded representations
|
|
encoding_dim = 32 # 32 floats -> compression of factor 24.5, assuming the input is 784 floats
|
|
sizes = [784, 400, 400, 10]
|
|
|
|
# this model maps an input to its reconstruction
|
|
coder = Model(Input(shape=(sizes[0],)),
|
|
Dense(sizes[1], activation='relu'),
|
|
Dense(sizes[2], activation='relu'),
|
|
Dense(sizes[3], activation='relu'),
|
|
Dense(sizes[4], activation='sigmoid')),
|
|
Dense(sizes[3], activation='relu'),
|
|
Dense(sizes[2], activation='relu'),
|
|
Dense(sizes[1], activation='relu'),
|
|
Dense(sizes[0], activation='sigmoid'))
|
|
|
|
|
|
decoder.fit()
|
|
|