test_rbm.py 文件源码

def test_fit_gibbs():
    # Gibbs on the RBM hidden layer should be able to recreate [[0], [1]]
    # from the same input
    rng = np.random.RandomState(42)
    X = np.array([[0.], [1.]])
    rbm1 = BernoulliRBM(n_components=2, batch_size=2,
                        n_iter=42, random_state=rng)
    # you need that much iters
    rbm1.fit(X)
    assert_almost_equal(rbm1.components_,
                        np.array([[0.02649814], [0.02009084]]), decimal=4)
    assert_almost_equal(rbm1.gibbs(X), X)
    return rbm1