Evaluate only unique xgrids #1587
Conversation
|
Actually, loading 100 times (order of number of datasets) the same 100 points xgrid/qgrid from 100 LHAPDF members doesn't seem to justify the added complication of caching the result, so probably it doesn't make sense to do it. I might play with it at some point in case it does improve things in real use cases. |
|
Greetings from your nice fit 🤖 !
Check the report carefully, and please buy me a ☕ , or better, a GPU 😉! |
scarlehoff
added
run-fit-bot
|
Greetings from your nice fit 🤖 !
Check the report carefully, and please buy me a ☕ , or better, a GPU 😉! |
|
Positivity seems ok (which is the only place where this would have an effect in the standard fit) https://vp.nnpdf.science/tmsUkZzYTfugL2uy5OoNfA== |
|
I'm not unhappy with the current version because it already works with some datases (i.e., positivity) of theory 200. So doesn't care about pineappl or not. If you guys are ok with it, it can be merged. If you still have doubts, we can wait as right now I have no idea about the gains that it would introduce as I need to be provided with smaller pineappl tables so that I can do a meaningful comparison. (in summary, I'm not dead set yet on this being useful beyond "oh cool, the pdf model is evaluated slightly faster") |
|
Greetings from your nice fit 🤖 !
Check the report carefully, and please buy me a ☕ , or better, a GPU 😉! |
Deals with #1586 in a very generic way so that one is still allow to mix situations in which not all xgrids are equal but some of them are.
For now only done for the fit. Once tests have run and it works there, I'll try to add some kind of cache to the vp PDF so that we can avoid calling
lhapdfonce per dataset (which I cannot promise because it is both trivial to do and trivial to do it in a disastrous way).