search

dfm / araa-gps

Souce code for our ARA&A review of Gaussian process regression for astronomical time-series
MIT License
28 stars 6 forks source link

Comparison of fitting transit with and without GP #18

Closed ericagol closed 1 year ago

ericagol commented 1 year ago

In Figure 3, it seems clear that the orange fit in the left panel is very poor - many of the error bars are 2-3 sigma away from the model. I'm wondering if a fairer comparison would be one in which the error bars are inflated (which has been common practice in exoplanet work), or detrending with a polynomial (also a common practice)? Otherwise the large chi-square causes a narrow posterior for this fit, I think, while I would expect that the discrepancy with the correct values should still remain.

https://github.com/dfm/araa-gps/blob/a659c2ccdcbd30bacf6bb52e845f071ebae08a1f/src/tex/ms.tex#L219-L226

dfm commented 1 year ago

Good point - I've added a jitter term to both models and it increases the uncertainties on the independent noise fit, but the results are still significantly inconsistent as expected. It might also be worth adding some words about model checking, but I've just left a note in the draft that we can revisit as necessary. (we're already in trouble because of page limits...)