Open dwysocki opened 9 years ago
One more thing, this also leaves open the possibility of passing an error function which is essentially a no-op, for when you only want a quickly fitted light curve, with no errors. Bootstrapping is going to be very time consuming, and so you probably only want to use it for publishable results.
We should consider how we want to handle computing errors from an API perspective. For Lasso regression, we will need to use bootstrapping, but for OLS, we can compute standard error. I propose we somehow provide an error function to
get_lightcurve
, or create a new function,get_lightcurve_with_error
, which takes that function. The bootstrapping function would need to callget_lightcurve
many times, while the standard error function would callget_lightcurve
a single time. Our current behavior could be recreated by passing in an error function which only computes the SDM of the light curve.Any ideas or concerns about this topic?