j-faria / bicerin Goto Github PK

There is a bug in the calculation of the Keplerian curves. We cycle through each component (planet) and then through each data point. The value of the Keplerian curve is calculated at that point and assigned to mu. Currently the assignment looks like this:

mu = background
for each planet
    for each data point
        mu = background + keplerian

which is obviously wrong! It should be

mu = background
for each planet
    for each data point
        mu += keplerian

This is a critical bug!!

The RV challenge data sets have ~500 points each. Without HODLR or another way to speed up the GPs, any analysis might be prohibitively slow...
I did some tests but I can't get the same values for the likelihood from HODLR and the simple normal inversion.