ae_finetuner.py 文件源码

def load_ae_encoder(path, nonlinearity=sigmoid):
    nn = sio.loadmat(path)
    w1 = nn['w1']
    w2 = nn['w2']
    w3 = nn['w3']
    w4 = nn['w4']
    b1 = nn['b1'][0]
    b2 = nn['b2'][0]
    b3 = nn['b3'][0]
    b4 = nn['b4'][0]

    layers = [
        (InputLayer, {'name': 'input', 'shape': (None, 1500)}),
        (DenseLayer, {'name': 'l1', 'num_units': 2000, 'nonlinearity': nonlinearity, 'W': w1, 'b': b1}),
        (DenseLayer, {'name': 'l2', 'num_units': 1000, 'nonlinearity': nonlinearity, 'W': w2, 'b': b2}),
        (DenseLayer, {'name': 'l3', 'num_units': 500, 'nonlinearity': nonlinearity, 'W': w3, 'b': b3}),
        (DenseLayer, {'name': 'l4', 'num_units': 50, 'nonlinearity': linear, 'W': w4, 'b': b4})
    ]

    '''
    dbn = NeuralNet(
        layers=layers,
        max_epochs=30,
        objective_loss_function=squared_error,
        update=nesterov_momentum,
        regression=True,
        verbose=1,
        update_learning_rate=0.001,
        update_momentum=0.05,
        objective_l2=0.005,
    )
    '''

    dbn = NeuralNet(
        layers=layers,
        max_epochs=10,
        objective_loss_function=squared_error,
        update=adadelta,
        regression=True,
        verbose=1,
        update_learning_rate=0.01,
        # update_learning_rate=0.001,
        # update_momentum=0.05,
        objective_l2=0.005,
    )
    return dbn