You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, waveforms are output with a shift in time compared to their EOB counterparts.
This was acceptable in earlier versions of the software, but it is not anymore, since we are trying to build a combined inspiral-postmerger model by hybridizing with NRPMw.
The approach is as follows: instead of computing the residuals in phase and flattening them by an arbitrary amount,
we almost flatten them by shifting the waveforms by an amount determined by a fit to the timeshifts at hand,
which was done by Matteo here.
So, the steps to take are:
include the time shift by the fit amount both in the generation of the training dataset (removing the flattening) and in the reconstruction
include the fit in mlgw-bns (function ParamsWithExtrinsic -> float)
test it
remove the possibility to shift in time when computing mismatches from existing tests (expecting a failure)
The text was updated successfully, but these errors were encountered:
Currently, waveforms are output with a shift in time compared to their EOB counterparts.
This was acceptable in earlier versions of the software, but it is not anymore, since we are trying to build a combined inspiral-postmerger model by hybridizing with
NRPMw
.The approach is as follows: instead of computing the residuals in phase and flattening them by an arbitrary amount,
we almost flatten them by shifting the waveforms by an amount determined by a fit to the timeshifts at hand,
which was done by Matteo here.
So, the steps to take are:
The text was updated successfully, but these errors were encountered: