Covariance Calculation methods seem to break

LSSTDESC / augur

DESC forecasting and inference validation tool

MIT License

4 stars 2 forks source link

Covariance Calculation methods seem to break #58

Open arthurmloureiro opened 4 months ago

arthurmloureiro commented 4 months ago

Methods like calling TJPCov and using the internal gaussian seem to be quite unstable leading to non-positive definite matrices and breaking the cholesky decomposition.

I haven't yet found a way to get a new covariance without causing the problem above.

bruno-moraes commented 4 months ago

Can confirm. I am having issues with 'gaus_internal' covariance generation in a LCDM + neutrinos run. I can provide a config_test.yml file if needed, but it should be easily reproducible.

arthurmloureiro commented 4 months ago

Tagging @felipeaoli here in case he has some insights on what is going on :D

felipeaoli commented 4 months ago

please, send us @bruno-moraes, share the config_test.yml. Thank you!