offline "rebalancing" : starts with samples that allow reweighting to any lam between 0 and 1, finds a sequence of lams such that heuristic_dist(lams[i], lams[i+1]) ~= threshold ( https://github.com/proteneer/timemachine/pull/538 )
This PR will extend the offline "rebalancing" functionality, with the intent of compressing the output of greedy bisection. (Motivation: bisection guarantees that the estimated overlap(lams[i], lams[i+1]) > threshold, but does not minimize len(lams).)
Currently TM has a few methods for adapting the number and spacing of lambda windows along a fixed path, depending on the setting:
This PR will extend the offline "rebalancing" functionality, with the intent of compressing the output of greedy bisection. (Motivation: bisection guarantees that the estimated overlap(lams[i], lams[i+1]) > threshold, but does not minimize len(lams).)