Adjust thresholds (#30)

* Lower threshold for checking after row-swap

* Relax check for stat. dist. invariance

Author: Marius1311

pygpcca/_gpcca.py

@@ -132,7 +132,7 @@ def _gram_schmidt_mod(X: np.ndarray, eta: np.ndarray) -> np.ndarray:
     X[:, 0] = np.sqrt(eta)
     # Raise, if the subspace changed!
     dummy = subspace_angles(X, Xc)
-    if not np.allclose(dummy, 0.0, atol=1e-8, rtol=1e-5):
+    if not np.allclose(dummy, 0.0, atol=1e-7, rtol=1e-5):
         logging.error(Xc)
         logging.error(X)
         raise ValueError(
@@ -153,7 +153,7 @@ def _gram_schmidt_mod(X: np.ndarray, eta: np.ndarray) -> np.ndarray:
 
     # Raise, if the subspace changed!
     dummy = subspace_angles(Q, Xc)
-    if not np.allclose(dummy, 0.0, atol=1e-8, rtol=1e-5):
+    if not np.allclose(dummy, 0.0, atol=1e-7, rtol=1e-5):
         raise ValueError(
             "The subspace of Q derived by eta-orthogonalization doesn't match the "
             f"subspace of the original Q! The subspace angles are: {dummy}. "

pygpcca/utils/_utils.py

@@ -416,7 +416,7 @@ def _is_stationary_distribution(T: Union[np.ndarray, spmatrix], pi: np.ndarray)
         raise ValueError("Shape mismatch.")
 
     # check for invariance
-    if not np.allclose(T.T.dot(pi), pi, rtol=1e4 * EPS, atol=1e4 * EPS):
+    if not np.allclose(T.T.dot(pi), pi, rtol=1e6 * EPS, atol=1e6 * EPS):
         dev = np.max(np.abs(T.T.dot(pi) - pi))
         raise ValueError(
             f"Stationary distribution is not invariant under the transition matrix. Maximal deviation = " f"{dev}"