def test_glmpca_pre_initialized(self):
f0 = np.random.randn(5, 2) / 10
l0 = np.random.randn(10, 2) / 10
g1 = glmpca.glmpca(self.Y,
2,
fam="poi",
init={
"factors": f0,
"loadings": l0
})
def run(self):
Y = self.data.X.todense().T
res = glmpca(Y, self.n_latent, fam=self.likelihood)
# Normalized matrix
norm = np.dot(res['factors'], res['loadings'].T)
self.data.obsm['X_norm'] = norm
self.data.obsm['X_emb'] = res['factors']
self.dump_to_h5ad(f"glmpca_{self.likelihood}")
def run_glmpca(counts,
fam,
theta=100,
penalty=1,
optimize_nb_theta=True,
maxIter=1000,
eps=0.0001,
n_PCs=50,
seed=42,
dataset_label=''):
'''Wrapper around GLM PCA by Will Townes: applies GLM PCA with given settings and saves results as pickle'''
np.random.seed(seed)
ctl = {"maxIter": maxIter, "eps": eps, "optimizeTheta": optimize_nb_theta}
if maxIter == 1000 and eps == 0.0001:
ctl_str = ''
else:
ctl_str = '_maxIter%u_eps%s' % (maxIter, eps)
starttime = str(datetime.now())
res = glmpca.glmpca(counts.T,
n_PCs,
fam=fam,
nb_theta=theta,
verbose=True,
penalty=penalty,
ctl=ctl)
endtime = str(datetime.now())
res['starttime'] = starttime
res['endtime'] = endtime
_ = res.pop('glmpca_family')
if fam == 'poi':
path = 'glmpca_results/%sglmpca-py_%s_penalty%u%s.pickle' % (
dataset_label, fam, penalty, ctl_str)
elif optimize_nb_theta:
path = 'glmpca_results/%sglmpca-py_%s_penalty%u%s.pickle' % (
dataset_label, fam, penalty, ctl_str)
else:
path = 'glmpca_results/%sglmpca-py_%s_fixedTheta%u_penalty%u%s.pickle' % (
dataset_label, fam, theta, penalty, ctl_str)
print('Saving at', path)
with open(path, 'wb') as f:
pickle.dump(res, f)
def test_glmpca_fixed_overdispersion(self):
fixed_theta = 123
g_nb= glmpca.glmpca(self.Y,2,fam="nb", nb_theta = fixed_theta, ctl={"maxIter":1000, "eps":1e-4, "optimizeTheta":False})
self.assertEqual(fixed_theta,g_nb['glmpca_family'].nb_theta)
def test_glmpca_extra_args(self):
g1= glmpca.glmpca(self.Y,2,fam="poi",verbose=True,penalty=10)
def test_glmpca_covariates(self):
X = np.array(range(1,6))
Z = np.array(range(1,11))
g1= glmpca.glmpca(self.Y,2,fam="poi",X=X[:,None],Z=Z[:,None])
def test_glmpca_dims_L1(self):
g1 = glmpca.glmpca(self.Y,1,fam="poi")
def test_glmpca_bern_likelihood(self):
g1 = glmpca.glmpca(self.Ybin,2,fam="bern",penalty=10)
def test_glmpca_mult_likelihood(self):
g1 = glmpca.glmpca(self.Y,2,fam="mult",sz=np.array(range(1,6)))
def test_glmpca_nb_likelihood(self):
g1 = glmpca.glmpca(self.Y,2,fam="nb")
def setUp(self):
self.Y = np.random.negative_binomial(4,.8,size=(10,5))
self.Ybin = self.Y.copy()
self.Ybin[self.Ybin>0] = 1
self.g1 = glmpca.glmpca(self.Y,2,fam="poi")
def run_glmpca(counts, n_latent=10, likelihood='poi'):
glm_res = glmpca(counts, n_latent, fam=likelihood)
glcpca_norm_counts = np.dot(glm_res['factors'], glm_res['loadings'].T)
return glcpca_norm_counts