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
The slowest part of the new halo model is currently the calculation of halo properties (stellar mass and SFR), these involve calculation of the power-law scaling relations and exponential turnovers for each halo.
There are two ways of speeding up this calculation I can think of:
Instead of storing halo mass, we store log(halo mass), which would remove all the pow() and exp() calls except for one at the end. We already mostly use log(M) in the sampling interpolation tables so it wouldn't require many changes and perhaps result in a slight speedup there as well. The only downside would be a slightly less intuitive output.
Since these relations are constant throughout a run, we can instead define interpolation tables for the mean scaling relations that only need to be initialised once, and tables mostly consisting of power-laws should be very accurate. This would likely be faster than the first option, but it would sacrifice a small amount accuracy near the turnovers and require the management of a few more interpolation tables.
The text was updated successfully, but these errors were encountered:
The slowest part of the new halo model is currently the calculation of halo properties (stellar mass and SFR), these involve calculation of the power-law scaling relations and exponential turnovers for each halo.
There are two ways of speeding up this calculation I can think of:
The text was updated successfully, but these errors were encountered: