def test_nmf_mask(simulate): x, eta = simulate mask = np.random.uniform(size=x.shape) < 0.25 y = np.ma.masked_array(x, mask=mask) res = nmf(y, 1) res0 = nmf(x, 1) assert (res != res0).any()
def imputation_score_nmf(x, rank):
try:
from wlra.nmf import nmf
res = nmf(x, rank, atol=1e-3)
return loss(res[x.mask], x.data[x.mask])
except RuntimeError:
return [np.nan for f in losses]
def generalization_score_nmf(train, test, rank=10, **kwargs):
from wlra.nmf import nmf
lam = nmf(train, rank=rank)
return pois_llik(lam, train, test)
def training_score_nmf(x, rank=10):
from wlra.nmf import nmf
return st.poisson(mu=nmf(x, rank)).logpmf(x).sum()
def test_nmf_shape(simulate): x, eta = simulate res = nmf(x, 1) assert res.shape == x.shape
def test_nmf_return_lf(simulate): x, eta = simulate l, f = nmf(x, 1, return_lf=True) assert l.shape == (x.shape[0], 1) assert f.shape == (1, x.shape[1])
def test_nmf_not_frob(simulate): x, eta = simulate res = nmf(x, 1, frob=False)
def test_nmf_objective(simulate): x, eta = simulate res = nmf(x, 1) obj = np.linalg.norm(x - res) res0 = skd.NMF(n_components=1).fit(x) assert np.isclose(obj, res0.reconstruction_err_)
def test_nmf_rank(simulate): x, eta = simulate res = nmf(x, 1) u, d, vt = np.linalg.svd(res) assert (~np.isclose(d, 0)).sum() == 1