ValueError: The subspace of Q derived by shifting a non-constant first (Schur)vector

[elgra]

Dear all,

since recently when trying to execute the following step:
g.fit(n_states=10, cluster_key="leiden")
g.plot_macrostates(which="all")
I'm running into the following error. Would you have any suggestions, what might cause this error? I was trying to find a possible reason for it but without success. Since this error pops out only since recently and since it's not the case when working with other data set, I assume the error is not due to the environment issues.
Any suggestions will be very much appreciated!

g.fit(n_states=10, cluster_key="leiden")

`ERROR:root:[[-2.74098831e-03 -7.92506849e-04 -4.62006579e-04 ... 5.48897836e-04
-4.46117253e-04 -5.87294109e-04]
[-2.74537506e-03 -4.22060822e-04 -5.62269660e-04 ... -1.11471560e-04
-8.69961090e-04 1.05912137e-03]
[-2.75787222e-03 -8.39866791e-04 6.06816683e-05 ... -5.34630419e-04
8.54067104e-05 -4.98222078e-04]
...
[-2.72375611e-03 2.86133521e-04 -1.37655462e-03 ... 6.87158514e-04
-1.88211525e-04 8.06324884e-05]
[-2.73241239e-03 -4.82351196e-04 -8.82360872e-04 ... 6.93345613e-04
9.29056292e-05 4.35279518e-04]
[-2.74020293e-03 -2.22056154e-04 -1.45526338e-03 ... 1.00860143e-03
6.22285670e-04 1.29442938e-03]]
ERROR:root:[[ 2.80307857e-03 -7.92506849e-04 -4.62006579e-04 ... 5.48897836e-04
-4.46117253e-04 -5.87294109e-04]
[ 2.80307857e-03 -4.22060822e-04 -5.62269660e-04 ... -1.11471560e-04
-8.69961090e-04 1.05912137e-03]
[ 2.80307857e-03 -8.39866791e-04 6.06816683e-05 ... -5.34630419e-04
8.54067104e-05 -4.98222078e-04]
...
[ 2.80307857e-03 2.86133521e-04 -1.37655462e-03 ... 6.87158514e-04
-1.88211525e-04 8.06324884e-05]
[ 2.80307857e-03 -4.82351196e-04 -8.82360872e-04 ... 6.93345613e-04
9.29056292e-05 4.35279518e-04]
[ 2.80307857e-03 -2.22056154e-04 -1.45526338e-03 ... 1.00860143e-03
6.22285670e-04 1.29442938e-03]]

ValueError Traceback (most recent call last)
Cell In[124], line 1
----> 1 g.fit(n_states=10, cluster_key="leiden")
2 g.plot_macrostates(which="all")

File ~/scratch/conda/envs/cr/lib/python3.10/site-packages/cellrank/estimators/terminal_states/_gpcca.py:563, in GPCCA.fit(self, n_states, n_cells, cluster_key, **kwargs)
560 n = n_states if isinstance(n_states, int) else max(n_states)
561 # call explicitly since compute_macrostates doesn't handle the case
562 # when minChi is used for n_states and self._gpcca is uninitialized
--> 563 _ = self.compute_schur(n, **kwargs)
564 return self.compute_macrostates(n_states=n_states, cluster_key=cluster_key, n_cells=n_cells)

File ~/scratch/conda/envs/cr/lib/python3.10/site-packages/cellrank/estimators/mixins/decomposition/_schur.py:175, in SchurMixin.compute_schur(self, n_components, initial_distribution, method, which, alpha, verbose)
173 with (contextlib.nullcontext if verbose else contextlib.redirect_stdout)(io.StringIO()):
174 try:
--> 175 self._gpcca._do_schur_helper(n_components)
176 except ValueError as e:
177 if "will split complex conjugate eigenvalues" not in str(e):

File ~/.local/lib/python3.10/site-packages/pygpcca/_gpcca.py:853, in GPCCA._do_schur_helper(self, m)
851 logging.info("Using pre-computed Schur decomposition")
852 else:
--> 853 self._p_X, self._p_R, self._p_eigenvalues = _do_schur(
854 self._P, eta=self._eta, m=m, z=self._z, method=self._method
855 )

File ~/.local/lib/python3.10/site-packages/pygpcca/_gpcca.py:260, in _do_schur(P, eta, m, z, method, tol_krylov)
258 if not np.allclose(Q.T.dot(Q * eta[:, None]), np.eye(Q.shape[1]), rtol=1e6 * EPS, atol=1e6 * EPS):
259 logging.debug("The Schur vectors aren't D-orthogonal so they are D-orthogonalized.")
--> 260 Q = _gram_schmidt_mod(Q, eta)
261 # Transform the orthonormalized Schur vectors of P_bar back
262 # to orthonormalized Schur vectors X of P.
263 X = np.true_divide(Q, np.sqrt(eta)[:, None])

File ~/.local/lib/python3.10/site-packages/pygpcca/_gpcca.py:138, in _gram_schmidt_mod(X, eta)
136 logging.error(Xc)
137 logging.error(X)
--> 138 raise ValueError(
139 "The subspace of Q derived by shifting a non-constant first (Schur)vector "
140 "to the right and setting the first (Schur) vector equal sqrt(eta) doesn't "
141 f"match the subspace of the original Q! The subspace angles are: {dummy}. "
142 f"Number of clusters: {m}."
143 )
145 # eta-orthonormalization
146 for j in range(m):

ValueError: The subspace of Q derived by shifting a non-constant first (Schur)vector to the right and setting the first (Schur) vector equal sqrt(eta) doesn't match the subspace of the original Q! The subspace angles are: [4.25185999e-02 1.14932390e-15 9.37777881e-16 8.45425796e-16
7.22915320e-16 6.54656692e-16 5.69987291e-16 4.62224178e-16
3.73019087e-16 3.32898212e-16]. Number of clusters: 10.`

---

cr.logging.print_versions()
cellrank==2.0.0 scanpy==1.9.3 anndata==0.9.2 numpy==1.24.4 numba==0.57.1 scipy==1.11.1 pandas==1.5.3 pygpcca==1.0.4 scikit-learn==1.3.0 statsmodels==0.14.0 python-igraph==0.10.8 scvelo==0.2.5 pygam==0.8.0 matplotlib==3.7.0 seaborn==0.12.2