auto-encoder.py

import lasagne
import numpy
import theano
import theano.tensor as T

def get_data():
    return numpy.random.randn(200, 32, 32), numpy.random.randint(2, size=(200, 1))

def build_ae(input_shape, input_var):
    l_in = lasagne.layers.InputLayer(shape=input_shape, input_var=input_var)

    # l_hid = lasagne.layers.Conv1DLayer(l_in, num_filters=32, filter_size=3,
    #                              nonlinearity=lasagne.nonlinearities.rectify,
    #                              W=lasagne.init.GlorotUniform())

    l_hid = lasagne.layers.Conv1DLayer(l_in, num_filters=32, filter_size=3, stride=2,
                                 nonlinearity=lasagne.nonlinearities.rectify,
                                 W=lasagne.init.GlorotUniform())
    
    l_out = lasagne.layers.InverseLayer(l_hid, l_hid)
    return l_out

def main():
    X, y = get_data()
    input_var = T.tensor3('inputs')

    data_shape = (None, X.shape[1], X.shape[2])
    network = build_ae(data_shape, input_var)

    prediction = lasagne.layers.get_output(network)
    loss = lasagne.objectives.squared_error(prediction, input_var)
    loss = loss.mean()

    params = lasagne.layers.get_all_params(network, trainable=True)
    updates = lasagne.updates.nesterov_momentum(loss, params, learning_rate=0.01, momentum=0.9)

if __name__ == "__main__":
    main()