lstm_classifier_baseline.py 文件源码

def create_model(input_shape, input_var, mask_shape, mask_var, lstm_size=250, output_classes=26,
                 w_init=las.init.Orthogonal()):
    gate_parameters = Gate(
        W_in=w_init, W_hid=w_init,
        b=las.init.Constant(0.))
    cell_parameters = Gate(
        W_in=w_init, W_hid=w_init,
        # Setting W_cell to None denotes that no cell connection will be used.
        W_cell=None, b=las.init.Constant(0.),
        # By convention, the cell nonlinearity is tanh in an LSTM.
        nonlinearity=tanh)

    l_in = InputLayer(input_shape, input_var, 'input')
    l_mask = InputLayer(mask_shape, mask_var, 'mask')

    f_lstm, b_lstm = create_blstm(l_in, l_mask, lstm_size, cell_parameters, gate_parameters, 'lstm')

    l_sum = ElemwiseSumLayer([f_lstm, b_lstm], name='sum')
    l_forward_slice1 = SliceLayer(l_sum, -1, 1, name='slice1')

    # Now, we can apply feed-forward layers as usual.
    # We want the network to predict a classification for the sequence,
    # so we'll use a the number of classes.
    l_out = DenseLayer(
        l_forward_slice1, num_units=output_classes, nonlinearity=las.nonlinearities.softmax, name='output')

    return l_out