Skip to content

Fix positivity-weights missmatch

Emanuele Roberto Nocera requested to merge reload_on_train_end into master

Created by: scarlehoff

This might fix the problems mentioned in #976 (closed), here's a quick example (I'm still running a full fit @wilsonmr, I should have ~50 replicas of the 70k fit by tomorrow). The problem seems to be solved (orange is before the fix).

matched_positivity_from_dataspecs11_dataspecs0_plot_positivity

WIP since the missing reload at the end of the fit could have an effect on the final fit (on the % of replicas that exhaust the training length)

Merge request reports