[keras] LSTM-Autoencoder internal result experiment

import tensorflow as tf

from tensorflow.keras.layers import Dense, LSTM, Masking, TimeDistributed, RepeatVector, Lambda, BatchNormalization
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.models import Model
from tensorflow.keras.models import Sequential

import numpy as np


def repeat_vector(args):
    """ Repeat vector n times """
    layer_to_repeat = args[0]
    sequence_layer = args[1]
    return RepeatVector(tf.keras.backend.shape(sequence_layer)[1])(layer_to_repeat)

#(samples，timesteps,features) samples=4，features=3, timesteps=variable length
train_X = np.array([
[[0, 1, 2], [9, 8, 7],[3, 6, 8]],
[[3, 4, 5]],
[[6, 7, 8], [6, 5, 4],[1, 7, 4]],
[[9, 0, 1], [3, 7, 4]]
])

test_X = np.array([
[[0, 1, 4]],
[[6, 7, 4], [7, 3, 0], [5, 8, 9], [0, 2, 4]]
])

train_Y = np.array([0, 1, 1, 0])

n_feat = 3

# padding
train_X = pad_sequences(train_X, padding='post')
test_X = pad_sequences(test_X, padding='post')

model = Sequential()

# masking to handle variable length size
inputs = tf.keras.Input(shape=(None, n_feat))
masked_input = Masking(mask_value=0)(inputs)

# encoder
encoder = LSTM(32, activation='tanh', return_sequences=True)(masked_input)
encoder = LSTM(units=16, activation='tanh', return_sequences=False)(encoder)

# decoder
decoder = Lambda(repeat_vector, output_shape=(None, n_feat)) ([encoder, masked_input])
decoder = LSTM(16, activation='tanh', return_sequences=True)(decoder)
decoder = LSTM(units=32, activation='tanh', return_sequences=True)(decoder)
decoder = TimeDistributed(Dense(units=n_feat))(decoder)

model = Model(inputs=inputs, outputs=decoder)
model.compile(optimizer='rmsprop', loss='mse')


mask_vector_layer = Model(inputs=model.inputs, outputs=model.layers[6].output)
print(mask_vector_layer.summary())
mask_vector_layer_output = mask_vector_layer.predict(test_X)
print(mask_vector_layer_output)