rodluger / starry_process

interpretable gaussian processes for stellar light curves
MIT License
16 stars 5 forks source link

Numerical issues at high SNR #2

Open rodluger opened 3 years ago

rodluger commented 3 years ago

When the error on the flux is < 1 ppt, inference breaks down. This can be seen in this test, which often fails (by a lot!) if ferr < 1e-3. Need to investigate this in more detail.