Method test_PCA

python/pytest/test_alg.py:418–501

()

    @staticmethod
    def test_PCA():
        np.random.seed(1)
        n = 1000
        p = 20
        s = 10
        group_size = 5
        group_num = 4
        support_size = np.zeros((p, 1))
        support_size[s - 1, 0] = 1

        x1 = np.random.randn(n, 1)
        x1 /= np.linalg.norm(x1)
        X = x1.dot(np.random.randn(1, p)) + 0.01 * np.random.randn(n, p)
        X = X - X.mean(axis=0)
        g_index = np.arange(group_num)
        g_index = g_index.repeat(group_size)

        # save_data(X, 'PCA')
        X = load_data('PCA')

        # null
        check_estimator(abess.SparsePCA())
        model1 = abess.SparsePCA(support_size=support_size)
        model1.fit(X)
        assert np.count_nonzero(model1.coef_) == s

        # ratio & transform
        model1.ratio(X)
        model1.transform(X)
        model1.fit_transform(X)

        # sparse
        model2 = abess.SparsePCA(support_size=s)
        model2.fit(coo_matrix(X), sparse_matrix=True)
        print("coef1: ", np.unique(np.nonzero(model1.coef_)[0]))
        print("coef2: ", np.unique(np.nonzero(model2.coef_)[0]))
        assert_value(model1.coef_, model2.coef_)

        model2 = abess.SparsePCA(support_size=s)
        model2.fit(X, sparse_matrix=True)
        assert_value(model1.coef_, model2.coef_)

        # sigma input
        model3 = abess.SparsePCA(support_size=support_size)
        model3.fit(Sigma=X.T.dot(X))
        model3.fit(Sigma=np.cov(X.T), n=n)
        assert_fit(model1.coef_, model3.coef_)

        # KPCA
        support_size_m = np.hstack((support_size, support_size, support_size))
        model4 = abess.SparsePCA(support_size=support_size_m)
        model4.fit(X, number=3)
        assert model4.coef_.shape[1] == 3

        for i in range(3):
            coef = np.nonzero(model4.coef_[:, i])[0]
            assert len(coef) == s