lattice/pyerrors
1
0
Fork 0
mirror of https://github.com/fjosw/pyerrors.git synced 2025-05-14 19:43:41 +02:00
Code Issues Projects Releases Packages Wiki Activity

docs: warning about ill conditioned hessian more detailed.

Browse source
This commit is contained in:
Fabian Joswig 2022-05-25 15:20:31 +01:00
parent 967ddb0ecd
commit 48636fedb2
1 changed files with 4 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

6
pyerrors/fits.py
View file

@ -269,7 +269,8 @@ def total_least_squares(x, y, func, silent=False, **kwargs):
raise Exception("It is required to use autograd.numpy instead of numpy within fit functions, see the documentation for details.") from None raise Exception("It is required to use autograd.numpy instead of numpy within fit functions, see the documentation for details.") from None
condn = np.linalg.cond(hess) condn = np.linalg.cond(hess)
if condn > 1e8: if condn > 1e8:
warnings.warn("Hessian matrix might be ill-conditioned ({0:1.2e}), error propagation might be unreliable.".format(condn), RuntimeWarning) warnings.warn("Hessian matrix might be ill-conditioned ({0:1.2e}), error propagation might be unreliable.\n \
Maybe try rescaling the problem such that all parameters are of O(1).".format(condn), RuntimeWarning)
try: try:
hess_inv = np.linalg.inv(hess) hess_inv = np.linalg.inv(hess)
except np.linalg.LinAlgError: except np.linalg.LinAlgError:
@ -556,7 +557,8 @@ def _standard_fit(x, y, func, silent=False, **kwargs):
raise Exception("It is required to use autograd.numpy instead of numpy within fit functions, see the documentation for details.") from None raise Exception("It is required to use autograd.numpy instead of numpy within fit functions, see the documentation for details.") from None
condn = np.linalg.cond(hess) condn = np.linalg.cond(hess)
if condn > 1e8: if condn > 1e8:
warnings.warn("Hessian matrix might be ill-conditioned ({0:1.2e}), error propagation might be unreliable.".format(condn), RuntimeWarning) warnings.warn("Hessian matrix might be ill-conditioned ({0:1.2e}), error propagation might be unreliable.\n \
Maybe try rescaling the problem such that all parameters are of O(1).".format(condn), RuntimeWarning)
try: try:
hess_inv = np.linalg.inv(hess) hess_inv = np.linalg.inv(hess)
except np.linalg.LinAlgError: except np.linalg.LinAlgError: