def test_pca__fit(sample_dataset):
est = pca.TruncatedSVD(compute=False)
est = pca.pca_fit(sample_dataset, est)
assert hasattr(est, "components_")
est = pca.TruncatedSVD(compute=False)
est = pca.pca_fit(sample_dataset.drop_vars("call_alternate_allele_count"), est)
assert hasattr(est, "components_")
def test_pca__transform(sample_dataset):
est = pca.TruncatedSVD()
est = pca.pca_fit(sample_dataset, est)
pca.pca_transform(sample_dataset, est).compute()
pca.pca_transform(
sample_dataset.drop_vars("call_alternate_allele_count"), est
).compute()
def test_pca__stats(sample_dataset):
# Test estimator with absent properties
est = pca.TruncatedSVD()
est = pca.pca_fit(sample_dataset, est)
del est.explained_variance_ratio_
ds = pca.pca_stats(xr.Dataset(), est)
assert "sample_pca_explained_variance_ratio" not in ds
assert "sample_pca_explained_variance" in ds
# Validate loadings by ensuring that the sum of squares
# across variants is equal to the eigenvalues (which are
# equal to the explained variance)
np.testing.assert_almost_equal(
(ds["sample_pca_loading"] ** 2).sum(dim="variants").values,
ds["sample_pca_explained_variance"].values,
decimal=3,
)
