Joost VandeVondele

@pb00068 has this exchange been concluded? The testing seems to suggest this could be removed, but you appeared to have a reservation.

I suggest we don't apply this PR for now, it is a bit to old to apply without repeated testing, and the risk of side-effects is too large.

What is a good way to test this patch? I tried just ``` setoption name nodestime value 500 ucinewgame go wtime 10000 ucinewgame quit ``` but all could be consistent,...

OK, testing match playing with fast-chess and parsing the pgn: ``` $ cat do_match_fastchess.sh ./fast-chess -rounds 1 -games 1 -tournament gauntlet -output format=cutechess -srand $RANDOM\ -concurrency 2 -openings file=UHO_Lichess_4852_v1.epd format=epd...

Some local testing confirms a nice gain: ``` Score of patch vs master: 552 - 450 - 998 [0.526] 2000 Elo difference: 17.73 +/- 10.78, LOS: 99.94 %, DrawRatio: 49.90...

@Sopel97 maybe you can test if this is fixed by adding that line to .travis.yml ?

@nodchip @Sopel97 this was auto-closed on merge, maybe we can explicitly add the line above to CI, to make sure it is fixed.

doesn't look like I can reopen @nodchip should have the permission to do that. github will auto-close issues if the commit message says `fixes #1234` unfortunately also for 'maybe fixes'...

this should be easy to add in `tests/instrumented_learn.sh`, we just need to generate a variant of learn01.exp ? Is the only change needed to set 'SkipLoadingEval false' ?

concerning loading older nets.. at least based on just the stockfish perspective, I think this is not so much of a problem. This is still 'eary days', so this kind...