def test_plra_oracle(simulate):
x, eta = simulate
l1 = st.poisson(mu=np.exp(
wlra.plra(x, rank=3, max_outer_iters=100,
check_converged=True))).logpmf(x).sum()
l0 = st.poisson(mu=np.exp(eta)).logpmf(x).sum()
assert l1 > l0
def imputation_score_plra1(x, rank):
try:
import wlra
res = np.exp(wlra.plra(x, rank=rank, max_outer_iters=1))
return loss(res[x.mask], x.data[x.mask])
except RuntimeError:
return [np.nan for f in losses]
def training_score_plra1(x, rank=10):
import wlra
lam = np.exp(wlra.plra(x, rank=rank))
return st.poisson(mu=lam).logpmf(x).sum()
def training_score_plra(x, rank):
import wlra
return st.poisson(mu=np.exp(
wlra.plra(x, rank=rank, max_outer_iters=100,
check_converged=True))).logpmf(x).sum()
def generalization_score_plra1(train, test, rank=10, **kwargs):
import wlra
lam = np.exp(wlra.plra(train, rank=rank))
return pois_llik(lam, train, test)
def test_plra1_10x():
import scmodes
x = scmodes.dataset.read_10x(
f'/project2/mstephens/aksarkar/projects/singlecell-ideas/data/10xgenomics/b_cells/filtered_matrices_mex/hg19/',
return_df=True)
res = wlra.plra(x.values, rank=10, verbose=True)
def test_plra_mask(simulate):
x, eta = simulate
mask = np.random.uniform(size=x.shape) < 0.25
x = np.ma.masked_array(x, mask=mask)
res = wlra.plra(x, 3)
def test_plra1_oracle(simulate):
x, eta = simulate
l1 = st.poisson(
mu=np.exp(wlra.plra(x, rank=3, max_outer_iters=1))).logpmf(x).sum()
l0 = st.poisson(mu=np.exp(eta)).logpmf(x).sum()
assert l1 > l0
def test_plra_assume_rank_1():
x = np.random.poisson(lam=np.exp(np.random.normal(size=(100, 200))))
res = wlra.plra(x, 1)
def test_plra_shape():
x = np.ones((100, 200))
res = wlra.plra(x, 1)
assert res.shape == (100, 200)
