"add variation" in analysis not always comply with blunder threshold
Chess Database and Toolkit program
Brought to you by:
stevenaaus
Menu
▾
▴
#60 "add variation" in analysis not always comply with blunder threshold
Hi!
This is a great software! I am only having a small issue when running the analysis. In my settings, I put a blunder threshold of 0.5, a cut-off threshold of 10 and I ask to include the variation when there is a blunder.
Despite these settings , sometimes the variation is not added when the the blunder is greater than 0.5. This mistake especially occurs when the evaluation is at around +/-5 or above, but I can also see it at lower levels.
Thanks
I use this feature a lot and have not found any reproducable bug. Pls post a game if you can reproduce an issue. In my experience, occasionally issues occur because of engine irregularities. If you are using stockfish, limiting its threads may help avoid it starving the gui of cpu.
Hmmm - I haven't changed the blunder variable in ages. Whyfor no one tells me there's a bug there. :(
Here is the game. I am using stockfish 15.1 depth 20. Blunder threshold 0.5 and cutoff threshold 10. Towards the end, at move 43 for white there is a jump from -5.81 to -6.83. Also in move 44 white, there is a jump from -6.36 to -7.08. These do not have variations.
Last edit: Ale 2023-05-11
Here is another example. Between moves 34 and 35 we go from an evaluation of -6.17 to -7.57 to -8.61 without any variation added.
Ok - after a quick stockfish 15.1 non-nnue install, and a single test, i cant reproduce this bug. (bug2 depth 20 , threshold .5 moves 34 and 35 have a 0.1 eval diff)
Perhaps you better supply me with an example with a simple unoptimised (or old) engine so we are more on the same page, as there are so many different SF compiles/options. I likr critter 1.6a from here https://www.vlasak.biz/critter/
Last edit: Steven 2023-05-12
bug2 pgn has this [Depth "22 , 26 , 20"] ???
I mean, it's quite possible scidvspc has a bug somewhere here, but i need a reproducible example, cheers.
Last edit: Steven 2023-05-12
Hi! Here it is. I downloaded the same game from chess.com, free of comments, and ran critter 1.6a 64 bit, depth 18. I set the threshold to 0.2 this time. You can find the same issue between moves 22-23, going from an evaluaiton of -3.73 to -4.05. Also between moves 27-28, going from evaluation -2.30 to -2.59.
I can't reproduce this on my box. One thing that comes to mind is it is possible that what you are seeing is a descrepancy due to rounding/formating to two decimal places. The actual score delta is not calculated on the rounded/displayed figures. But i don't even see this really.
Perhaps have a study of your engines log. But the descrepancies you report are so marginal....
Hi, thanks for looking it up. I don't know what else to say :) Anyway, this issue keeps occurring every now and then, and not only with marginal errors. If you are still willing to give it another go, here I have another game where I made a lot of silly mistakes, so the evalutaion keeps oscillating a lot. There are only 12 moves. Here I have issue with stockfish 15.1 and also with the pre-loaded stockfish.
Despite this issue, I still think that this software is one of the best free tools for chess analysis :)
Ok, i tested this pgn on my old macbook / critter / 0.5 thresh/ 15.0 cutoff, and it went perfectly.
I know there are issues related to engine not getting enough CPU, and the engine PVs are a little confusing. Maybe i'll look harder at it one day,
Stockfish is a CPU hogging beast which i try not to use. Our poor scid GUI runs interpreted! wish.