test_rbm.py 文件源码

def test_fit_gibbs_sparse():
    # Gibbs on the RBM hidden layer should be able to recreate [[0], [1]] from
    # the same input even when the input is sparse, and test against non-sparse
    rbm1 = test_fit_gibbs()
    rng = np.random.RandomState(42)
    from scipy.sparse import csc_matrix
    X = csc_matrix([[0.], [1.]])
    rbm2 = BernoulliRBM(n_components=2, batch_size=2,
                        n_iter=42, random_state=rng)
    rbm2.fit(X)
    assert_almost_equal(rbm2.components_,
                        np.array([[0.02649814], [0.02009084]]), decimal=4)
    assert_almost_equal(rbm2.gibbs(X), X.toarray())
    assert_almost_equal(rbm1.components_, rbm2.components_)