Training the autoencoder model with smaller number of timesteps

Study the impact of the encoder size with reconstruction error of the model. Try with timesteps of lengths: 20, 50, 100, 300, 500, 1000.