test_models.py 文件源码

def test_model_custom_loss():
    x = torch.rand(20, 4)
    y = torch.rand(20, 10)

    model = Model(
        Dense(10, input_dim=x.size()[-1]),
        Activation('relu'),
        Dense(5),
        Activation('relu'),
        Dense(y.size()[-1])
    )

    opt = SGD(lr=0.01, momentum=0.9)

    def mae(y_true, y_pred):
        return torch.mean(torch.abs(y_true - y_pred))

    history = model.fit(x, y, loss=mae, optimizer=opt, epochs=10)
    assert len(history['loss']) == 10
    assert all(type(v) is float for v in history['loss'])
    assert history['loss'] == sorted(history['loss'], reverse=True)