GlmpcaError: Numerical divergence (deviance no longer finite), try increasing the penalty to improve stability of optimization. #7
Description
Hi,
I am interested in this tool for constructing 3D model of Stere-seq datasets. An error is raised when I run partial_pairwise_align. I'm not quite sure if the data is too large.
Thank you!
/mnt/data1/robot/software/miniconda3/envs/paste2/lib/python3.9/site-packages/statsmodels/genmod/families/links.py:527: RuntimeWarning: overflow encountered in exp
return np.exp(z)
/mnt/data1/robot/software/miniconda3/envs/paste2/lib/python3.9/site-packages/paste2/glmpca.py:393: RuntimeWarning: invalid value encountered in divide
U[:, k] += grads / infos
/mnt/data1/robot/software/miniconda3/envs/paste2/lib/python3.9/site-packages/statsmodels/genmod/families/family.py:445: RuntimeWarning: invalid value encountered in divide
endog_mu = self._clean(endog / mu)
---------------------------------------------------------------------------
GlmpcaError Traceback (most recent call last)
Cell In[27], line 1
----> 1 pi_AB = PASTE2.partial_pairwise_align(adata_st_lst[0], adata_st_lst[1], s=0.7)
File /mnt/data1/robot/software/miniconda3/envs/paste2/lib/python3.9/site-packages/paste2/PASTE2.py:260, in partial_pairwise_align(sliceA, sliceB, s, alpha, armijo, dissimilarity, use_rep, G_init, a_distribution, b_distribution, norm, return_obj, verbose)
258 M = pca_distance(sliceA, sliceB, 2000, 20)
259 elif dissimilarity.lower() == 'glmpca':
--> 260 M = glmpca_distance(A_X, B_X, latent_dim=50, filter=True, verbose=verbose)
261 else:
262 print("ERROR")
File /mnt/data1/robot/software/miniconda3/envs/paste2/lib/python3.9/site-packages/paste2/helper.py:82, in glmpca_distance(X, Y, latent_dim, filter, verbose)
79 joint_matrix = joint_matrix[:, top_indices]
81 print("Starting GLM-PCA...")
---> 82 res = glmpca(joint_matrix.T, latent_dim, penalty=1, verbose=verbose)
83 #res = glmpca(joint_matrix.T, latent_dim, fam='nb', penalty=1, verbose=True)
84 reduced_joint_matrix = res["factors"]
File /mnt/data1/robot/software/miniconda3/envs/paste2/lib/python3.9/site-packages/paste2/glmpca.py:374, in glmpca(Y, L, fam, ctl, penalty, verbose, init, nb_theta, X, Z, sz)
372 dev[t] = gf.dev_func(Y, rfunc(U, V))
373 if not np.isfinite(dev[t]):
--> 374 raise GlmpcaError(
375 "Numerical divergence (deviance no longer finite), try increasing the penalty to improve stability of optimization.")
376 if t > 4 and np.abs(dev[t] - dev[t - 1]) / (0.1 + np.abs(dev[t - 1])) < ctl["eps"]:
377 break
GlmpcaError: Numerical divergence (deviance no longer finite), try increasing the penalty to improve stability of optimization.