Rebel wrote: ↑Sun Oct 23, 2022 8:37 am
I want people to give a choice which engine they trust most. I am definitely in favor of Lc0.
I'd say it depends on the position, but if I do favour LC0's choice over SF's now and then too, (most of the time with enough hardware- time they become the same anyhow, not the evals, but the moves of first choice) )then I choose LC0's move. If I yet couldn't decide for one of two moves (often enough they transpose sooner or later e.g.), then I'd give same amount of points to both moves,
regards
Ferdy wrote: ↑Mon Oct 24, 2022 4:43 am
There is sts positions analyzed by sf15 at 60s with multipv 10 on a single core using cpu i7-2600K cpu (3.4 Ghz).
EPD : epd\sts-sf15-ferdy.epd
Time : 100ms
Max Total Time Hash
Engine Score Found Pos ELO Score Rate ms Mb Cpu
1 Berserk 9 13452 1025 1500 4041 14973 89.8% 100 64 1
2 Koivisto 8.0 13436 1037 1500 4036 14973 89.7% 100 64 1
3 Rebel 15.1 13237 1001 1500 3978 14973 88.4% 100 64 1
4 Seer 2.5.0 12985 957 1500 3901 14973 86.7% 100 64 1
5 Toga 4 11575 753 1500 3478 14973 77.3% 100 64 1
6 Fruit 2.1 11017 678 1500 3312 14973 73.6% 100 64 1
Created with MEA
by
Ferdinand
Mosca
My move bonus point calculation is a bit different than yours. First move is always 10, for remaining moves the score difference with the first move is divided by 5. Moves with a score difference is > 50 are not listed.
For @Srdja, the Ferdy EPD also contains long moves, g1f3 instead of Nf3.
90% of coding is debugging, the other 10% is writing bugs.
Ferdy wrote: ↑Mon Oct 24, 2022 4:43 am
There is sts positions analyzed by sf15 at 60s with multipv 10 on a single core using cpu i7-2600K cpu (3.4 Ghz).
EPD : epd\sts-sf15-ferdy.epd
Time : 100ms
Max Total Time Hash
Engine Score Found Pos ELO Score Rate ms Mb Cpu
1 Berserk 9 13452 1025 1500 4041 14973 89.8% 100 64 1
2 Koivisto 8.0 13436 1037 1500 4036 14973 89.7% 100 64 1
3 Rebel 15.1 13237 1001 1500 3978 14973 88.4% 100 64 1
4 Seer 2.5.0 12985 957 1500 3901 14973 86.7% 100 64 1
5 Toga 4 11575 753 1500 3478 14973 77.3% 100 64 1
6 Fruit 2.1 11017 678 1500 3312 14973 73.6% 100 64 1
Created with MEA
by
Ferdinand
Mosca
My move bonus point calculation is a bit different than yours. First move is always 10, for remaining moves the score difference with the first move is divided by 5. Moves with a score difference is > 50 are not listed.
For @Srdja, the Ferdy EPD also contains long moves, g1f3 instead of Nf3.
I have added STS1-STS15_LAN_v5.epd. Its max point is 100 instead of 10. The weakness of max 10-point system that I see in STS1-STS15_LAN_v4.epd is that the range is too narrow and it has difficulties differentiating engines that are close in strength.
I describe the interpolation that I use in the repository.
I reran the 1497 distinct positions at 1 minute per position using multi-pv using two Nvidia 2080 Super GPUs and LC0-29.2 with the 300MB weights file:
d0ed346c32fbcc9eb2f0bc7e957d188c8ae428ee3ef7291fd5aa045fc6ef4ded
The results are found in this file:
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
