https://roadmap.strategyquant.com/tasks/sq4_7523
I've been watching the process carefully and reading the related tickets.
What's actually happening with the LOW CPU USAGE is that during the initial fill stage when no islands are full yet CPU is still using 100% BUT the islands take 10x as long to fill up (because of a double whammy: there are false positives as I've shown in my screenshot above AND because the correlation filter is incorrectly used as a init pop filter instead of a ranking filter.) CPU usage does eventually drop to 50% when only one or two or three islands are actually full (which, once the first island actually finally fills, is often. It might take 30 minutes or 60 minutes to get there but it drops to 50% once the first island is completely full)
Subject changed from Fit to existing portfolio cannot possibly be right to Fit to existing portfolio cannot possibly be right: two bugs found
BUG #1 FALSE POSITIVES
correlation itself is calculated correctly.
In the attached config file the option "Allow negative correlation" is off.
I assume that this is a reason why your strategies are dismissed.
Check Engine log for messages like - Fit to existing portfolio - dismissed: corr_test correlation -0.13 < 0
BUG #2 CORRELATION FILTER IS AN INIT POP FILTER INSTEAD OF RANKING FILTER
thanks for pointing it out, fixed
If turned OFF, it removes all strategies whose correlation is > 0.3 or < 0
If turned ON, it removes all strategies whose correlation is > 0.3, and keeps strategies with any negative correlation.
Perhaps it should be set ON on default.
my entire portfolio of one strategy is attached. I'd be surprised if another strategy didn't produce the same or similar results though...