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, reweighting works pretty well, apart from cases like this:
We end up "spending" the observed GA's on the first few rounds of path finding. Leaving us only able to select the unweighted GG option in future paths. This is technically correct, but we need a method that perhaps uses more delicate reweighting?
The text was updated successfully, but these errors were encountered:
In a quick and dirty fashion, we can cap the ratio used to reweight the Hansel matrix and this yields more accurate results (on the order of 0.2-2%). This somewhat confirms the hypothesis that aggressive reweighting is likely to be the cause of the close-but-no-cigar results on some of our datasets.
Currently, reweighting works pretty well, apart from cases like this:
We end up "spending" the observed GA's on the first few rounds of path finding. Leaving us only able to select the unweighted GG option in future paths. This is technically correct, but we need a method that perhaps uses more delicate reweighting?
The text was updated successfully, but these errors were encountered: