Knight value

[Marcus9]

I have played Dragon 2.6 against itself a with 300 deltaelo since 2300, so 2300 played 1000 games against 2000, 2400 played 1000 games against 2100 .. and so on. Stronger side always played knight odd against weaker side, I used ChrisW's nide-odds-masterlist-1000. The strongest pair played was 3000 vs 2700.

https://ibb.co/YpMK8gs

Adding the difference elo of 300 we can extrapolate the knight value (I omit '00' for the sake of brevity):
23 vs 20: 243.6
24 vs 21: 249.3
25 vs 22: 221.2
26 vs 23: 189.8
27 vs 24: 250.7
28 vs 25: 322.2
29 vs 26: 412.9
30 vs 27: 459.7

I expect this to be accurate enough to gauge a knight's worth in engine vs engine matches, however we do not see an ever increasing trend in the lower elo range (why?).
I did this test to extrapolate the value of a knight to higher elo with an appropiate regression.
However, I don't know how accurate the differences in komodo skill levels are.
In the future I will perform similar tests with different deltaelo.

[Marcus9]

I have played Dragon 2.6 against itself a with 300 deltaelo since 2300, so 2300 played 1000 games against 2000, 2400 played 1000 games against 2100 .. and so on. Stronger side always played knight odd against weaker side, I used ChrisW's nide-odds-masterlist-1000. The strongest pair played was 3000 vs 2700.

https://ibb.co/YpMK8gs

Adding the difference elo of 300 we can extrapolate the knight value (I omit '00' for the sake of brevity):
23 vs 20: 243.6
24 vs 21: 249.3
25 vs 22: 221.2
26 vs 23: 189.8
27 vs 24: 250.7
28 vs 25: 322.2
29 vs 26: 412.9
30 vs 27: 459.7

I expect this to be accurate enough to gauge a knight's worth in engine vs engine matches, however we do not see an ever increasing trend in the lower elo range (why?).
I did this test to extrapolate the value of a knight to higher elo with an appropiate regression.
However, I don't know how accurate the differences in komodo skill levels are.
In the future I will perform similar tests with different deltaelo.

* 'variety' is set to 10 in all games

[lkaufman]

I have played Dragon 2.6 against itself a with 300 deltaelo since 2300, so 2300 played 1000 games against 2000, 2400 played 1000 games against 2100 .. and so on. Stronger side always played knight odd against weaker side, I used ChrisW's nide-odds-masterlist-1000. The strongest pair played was 3000 vs 2700.

https://ibb.co/YpMK8gs

Adding the difference elo of 300 we can extrapolate the knight value (I omit '00' for the sake of brevity):
23 vs 20: 243.6
24 vs 21: 249.3
25 vs 22: 221.2
26 vs 23: 189.8
27 vs 24: 250.7
28 vs 25: 322.2
29 vs 26: 412.9
30 vs 27: 459.7

I expect this to be accurate enough to gauge a knight's worth in engine vs engine matches, however we do not see an ever increasing trend in the lower elo range (why?).
I did this test to extrapolate the value of a knight to higher elo with an appropiate regression.
However, I don't know how accurate the differences in komodo skill levels are.
In the future I will perform similar tests with different deltaelo.

* 'variety' is set to 10 in all games

These values for knight odds seem ridiculously low; many tests I have run (with the odds receiver around 2500 level) put the value around 900 to 1000 elo, using a variety of methods. I'm not sure what the problem is, but I suggest trying again when Dragon 3 comes out (which won't be long), as it will have Elo values much more accurately calibrated to be like human FIDE ratings if they are playing 15' + 10" Rapid. Also you should not set "Variety"; the Elo settings each have their own variety values, and they are larger than 10 even in GM range. In human terms, it is absurd to think that a 2800 FIDE GM (i.e. roughly World 2-5) could give knight odds in Rapid to a typical 2500 FIDE GM and score anywhere near even. Whether the problem is that the elo values in Dragon 2.6 were not so accurate or just that engine chess is not like human chess is hard to say, undoubtedly both factors contribute to your result. But even allowing for these factors I find the results quite puzzling.

[Marcus9]

I have played Dragon 2.6 against itself a with 300 deltaelo since 2300, so 2300 played 1000 games against 2000, 2400 played 1000 games against 2100 .. and so on. Stronger side always played knight odd against weaker side, I used ChrisW's nide-odds-masterlist-1000. The strongest pair played was 3000 vs 2700.

https://ibb.co/YpMK8gs

Adding the difference elo of 300 we can extrapolate the knight value (I omit '00' for the sake of brevity):
23 vs 20: 243.6
24 vs 21: 249.3
25 vs 22: 221.2
26 vs 23: 189.8
27 vs 24: 250.7
28 vs 25: 322.2
29 vs 26: 412.9
30 vs 27: 459.7

I expect this to be accurate enough to gauge a knight's worth in engine vs engine matches, however we do not see an ever increasing trend in the lower elo range (why?).
I did this test to extrapolate the value of a knight to higher elo with an appropiate regression.
However, I don't know how accurate the differences in komodo skill levels are.
In the future I will perform similar tests with different deltaelo.

* 'variety' is set to 10 in all games

These values for knight odds seem ridiculously low; many tests I have run (with the odds receiver around 2500 level) put the value around 900 to 1000 elo, using a variety of methods. I'm not sure what the problem is, but I suggest trying again when Dragon 3 comes out (which won't be long), as it will have Elo values much more accurately calibrated to be like human FIDE ratings if they are playing 15' + 10" Rapid. Also you should not set "Variety"; the Elo settings each have their own variety values, and they are larger than 10 even in GM range. In human terms, it is absurd to think that a 2800 FIDE GM (i.e. roughly World 2-5) could give knight odds in Rapid to a typical 2500 FIDE GM and score anywhere near even. Whether the problem is that the elo values in Dragon 2.6 were not so accurate or just that engine chess is not like human chess is hard to say, undoubtedly both factors contribute to your result. But even allowing for these factors I find the results quite puzzling.

Effectively I got very low values ​​and the speed of my dragon's moves seem faster, therefore weaker, this would explain why the knight has such a low value since I am having lower strength engines challenged than I expect . The problem is probably me, I'll investigate this.
Meanwhile can someone confirm that Dragon2.6 vs Dragon2.6 setting at 2000 use (or not) about half second/game without GUI?

[Marcus9]

I have played Dragon 2.6 against itself a with 300 deltaelo since 2300, so 2300 played 1000 games against 2000, 2400 played 1000 games against 2100 .. and so on. Stronger side always played knight odd against weaker side, I used ChrisW's nide-odds-masterlist-1000. The strongest pair played was 3000 vs 2700.

https://ibb.co/YpMK8gs

Adding the difference elo of 300 we can extrapolate the knight value (I omit '00' for the sake of brevity):
23 vs 20: 243.6
24 vs 21: 249.3
25 vs 22: 221.2
26 vs 23: 189.8
27 vs 24: 250.7
28 vs 25: 322.2
29 vs 26: 412.9
30 vs 27: 459.7

I expect this to be accurate enough to gauge a knight's worth in engine vs engine matches, however we do not see an ever increasing trend in the lower elo range (why?).
I did this test to extrapolate the value of a knight to higher elo with an appropiate regression.
However, I don't know how accurate the differences in komodo skill levels are.
In the future I will perform similar tests with different deltaelo.

* 'variety' is set to 10 in all games

These values for knight odds seem ridiculously low; many tests I have run (with the odds receiver around 2500 level) put the value around 900 to 1000 elo, using a variety of methods. I'm not sure what the problem is, but I suggest trying again when Dragon 3 comes out (which won't be long), as it will have Elo values much more accurately calibrated to be like human FIDE ratings if they are playing 15' + 10" Rapid. Also you should not set "Variety"; the Elo settings each have their own variety values, and they are larger than 10 even in GM range. In human terms, it is absurd to think that a 2800 FIDE GM (i.e. roughly World 2-5) could give knight odds in Rapid to a typical 2500 FIDE GM and score anywhere near even. Whether the problem is that the elo values in Dragon 2.6 were not so accurate or just that engine chess is not like human chess is hard to say, undoubtedly both factors contribute to your result. But even allowing for these factors I find the results quite puzzling.

Effectively I got very low values ​​and the speed of my dragon's moves seem faster, therefore weaker, this would explain why the knight has such a low value since I am having lower strength engines challenged than I expect . The problem is probably me, I'll investigate this.
Meanwhile can someone confirm that Dragon2.6 vs Dragon2.6 setting at 2000 use (or not) about half second/game without GUI?

I think I found the bug, some draws were counted as wins. I'm running the same match again

[Marcus9]

I have played Dragon 2.6 against itself a with 300 deltaelo since 2300, so 2300 played 1000 games against 2000, 2400 played 1000 games against 2100 .. and so on. Stronger side always played knight odd against weaker side, I used ChrisW's nide-odds-masterlist-1000. The strongest pair played was 3000 vs 2700.

https://ibb.co/YpMK8gs

Adding the difference elo of 300 we can extrapolate the knight value (I omit '00' for the sake of brevity):
23 vs 20: 243.6
24 vs 21: 249.3
25 vs 22: 221.2
26 vs 23: 189.8
27 vs 24: 250.7
28 vs 25: 322.2
29 vs 26: 412.9
30 vs 27: 459.7

I expect this to be accurate enough to gauge a knight's worth in engine vs engine matches, however we do not see an ever increasing trend in the lower elo range (why?).
I did this test to extrapolate the value of a knight to higher elo with an appropiate regression.
However, I don't know how accurate the differences in komodo skill levels are.
In the future I will perform similar tests with different deltaelo.

* 'variety' is set to 10 in all games

These values for knight odds seem ridiculously low; many tests I have run (with the odds receiver around 2500 level) put the value around 900 to 1000 elo, using a variety of methods. I'm not sure what the problem is, but I suggest trying again when Dragon 3 comes out (which won't be long), as it will have Elo values much more accurately calibrated to be like human FIDE ratings if they are playing 15' + 10" Rapid. Also you should not set "Variety"; the Elo settings each have their own variety values, and they are larger than 10 even in GM range. In human terms, it is absurd to think that a 2800 FIDE GM (i.e. roughly World 2-5) could give knight odds in Rapid to a typical 2500 FIDE GM and score anywhere near even. Whether the problem is that the elo values in Dragon 2.6 were not so accurate or just that engine chess is not like human chess is hard to say, undoubtedly both factors contribute to your result. But even allowing for these factors I find the results quite puzzling.

Effectively I got very low values ​​and the speed of my dragon's moves seem faster, therefore weaker, this would explain why the knight has such a low value since I am having lower strength engines challenged than I expect . The problem is probably me, I'll investigate this.
Meanwhile can someone confirm that Dragon2.6 vs Dragon2.6 setting at 2000 use (or not) about half second/game without GUI?

I think I found the bug, some draws were counted as wins. I'm running the same match again

I repeated the test with the same result:

2300 elo vs 2000 elo:
WDL: [434, 275, 291]
elo performance: 50.02616338883343 ± 18.471133880884636
Finished in: 560.3892095088959 seconds

2400 elo vs 2100 elo:
WDL: [463, 260, 277]
elo performance: 65.384113655617 ± 18.75450397468894
Finished in: 682.4239001274109 seconds

2500 elo vs 2200 elo:
WDL: [476, 248, 276]
elo performance: 70.43650362227247 ± 18.948775563045732
Finished in: 850.0434765815735 seconds

2600 elo vs 2300 elo:
WDL: [501, 266, 233]
elo performance: 95.44326899492881 ± 18.901981017845976
Finished in: 1130.1557455062866 seconds

2700 elo vs 2400 elo:
WDL: [414, 253, 333]
elo performance: 28.204073026879616 ± 18.670666329210327
Finished in: 1626.627239704132 seconds

2800 elo vs 2500 elo:
WDL: [309, 262, 429]
elo performance: -41.89414020800519 ± 18.604230429156093
Finished in: 2534.6143085956573 seconds

2900 elo vs 2600 elo:
WDL: [239, 238, 523]
elo performance: -101.46080056999156 ± 19.359472344577057
Finished in: 3916.948734521866 seconds

3000 elo vs 2700 elo:
WDL: [122, 277, 601]
elo performance: -181.25218027894718 ± 19.6687583228414
Finished in: 6684.236461162567 seconds

Total time: 17985.43907570839 seconds