Stockfish wins TCEC

[lkaufman]

Larry,

it is not due to better time management. It is because of much better tactical skills. Try to test Komodo vs Stockfish in solving tactical puzzles...

I know that SF is better at tactics, and I think that this is offset by positional superiority in Komodo. Time management can account for most of SF elo lead. I base this on tests I ran at fast time controls that show that with pure increment time limit (for example 0.3" plus 0.3" per move) we are within a few elo of SF now, but with normal time limits (where base time is 100 or more times the increment) the gap is in the twenties. To be precise, I measured an 18 elo difference between these two types of time limits (keeping average time and strength about the same).
Of course you could also say that tactics are the margin of superiority, with position advantage of Komodo offset by inferior time management. That would also be accurate.
So we could improve Komodo by improving tactical skill or by improving time management. But the second of these is probably easier to do. Stockfish doesn't have much code that helps its tactics, so I assume that its tactical edge is due to heavier pruning. But if we do that we get weaker overall.

[M ANSARI]

I think SF is stronger because it reaches deep into a position much quicker. I did not feel that Komodo's TM was poor, but really the longer the time control the less critical time management is. Of course that doesn't mean that any poor TM will work, but simply that it is not as critical. If you look at Komodo's losses, they were simply due to not being able to see as deep into the position as SF. At this level, being a few ply shallower means that the chances of being outplayed will be higher even if you might have a more accurate evaluation. An accurate evaluation only works if there are no tactical tricks to refute the positional evaluation.

I do think that SF has set the bar very high, but am pretty sure Komodo will be vying for the top spot for a while yet. By the way great job with Komodo and the final game was a real gem!

[Uri Blass]

I think that time managament is not the only reason that stockfish is stronger,but I think that komodo played the first moves too fast

Here are the times that the programs had in the clock for the rest of the moves after move 30(first number is stockfish's time).

You can see that in most cases stockfish had less time on the clock after move 30 or in the end of the game if the end happened earlier
(in 54 cases out of 64 cases).

In 19 cases the time gap was more than 20 minutes if I count correctly and in all of them stockfish used more time.


1) 53:42 58:27
2) 61:50 76:59
3) 77:52 89:15 29 moves
4) 88:16 75:41 22 moves
5) 71:11 89:20
6) 38:48 62:20
7) 54:50 51:55
8) 57:55 70:55
9) 53:39 67:21
10)49:21 69:47
11)56:54 81:28
12)49:00 72:38
13)67:27 73:36
14)68:30 62:59
15)45:13 78:56
16)55:18 58:51
17)36:05 71:43
18)45:54 44:57
19)63:18 61:41
20)56:56 78:02
21)62:17 78:16
22)56:04 68:36
23)54:33 71:19
24)39:21 66:01
25)71:04 76:10
26)72:19 78:51
27)41:53 77:28
28)28:23 90:11
29)57:14 73:43
30)55:26 68:33
31)68:54 59:31
32)59:02 69:55
33)53:47 72:30
34)58:03 63:58
35)69:55 74:05
36)60:17 64:44
37)60:05 75:49
38)68:17 80:11
39)32:01 65:17
40)64:02 50:13
41)38:11 55:29
42)65:23 57:39
43)26:33 56:51
44)35:07 64:47
45)68:37 61:27
46)53:21 54:21
47)42:35 44:06
48)39:30 70:06
49)56:39 88:54
50)64:14 73:13
51)49:32 67:06
52)46:36 69:56
53)63:25 78:52
54)55:49 70:16
55)45:17 77:11
56)67:03 68:53
57)93:40 78:23 22 moves
58)59:35 73:48
59)56:46 77:59
60)59:09 74:54
61)61:21 76:35
62)63:39 77:10
63)50:17 62:07
64)44:48 80:29

[Uri Blass]

I think SF is stronger because it reaches deep into a position much quicker. I did not feel that Komodo's TM was poor, but really the longer the time control the less critical time management is. Of course that doesn't mean that any poor TM will work, but simply that it is not as critical. If you look at Komodo's losses, they were simply due to not being able to see as deep into the position as SF.

Part of the reason that komodo did not see deep enough is because it used less time than stockfish in the first moves.

[lkaufman]

I think that time managament is not the only reason that stockfish is stronger,but I think that komodo played the first moves too fast

Here are the times that the programs had in the clock for the rest of the moves after move 30(first number is stockfish's time).

You can see that in most cases stockfish had less time on the clock after move 30 or in the end of the game if the end happened earlier
(in 54 cases out of 64 cases).

In 19 cases the time gap was more than 20 minutes if I count correctly and in all of them stockfish used more time.


1) 53:42 58:27
2) 61:50 76:59
3) 77:52 89:15 29 moves
4) 88:16 75:41 22 moves
5) 71:11 89:20
6) 38:48 62:20
7) 54:50 51:55
8) 57:55 70:55
9) 53:39 67:21
10)49:21 69:47
11)56:54 81:28
12)49:00 72:38
13)67:27 73:36
14)68:30 62:59
15)45:13 78:56
16)55:18 58:51
17)36:05 71:43
18)45:54 44:57
19)63:18 61:41
20)56:56 78:02
21)62:17 78:16
22)56:04 68:36
23)54:33 71:19
24)39:21 66:01
25)71:04 76:10
26)72:19 78:51
27)41:53 77:28
28)28:23 90:11
29)57:14 73:43
30)55:26 68:33
31)68:54 59:31
32)59:02 69:55
33)53:47 72:30
34)58:03 63:58
35)69:55 74:05
36)60:17 64:44
37)60:05 75:49
38)68:17 80:11
39)32:01 65:17
40)64:02 50:13
41)38:11 55:29
42)65:23 57:39
43)26:33 56:51
44)35:07 64:47
45)68:37 61:27
46)53:21 54:21
47)42:35 44:06
48)39:30 70:06
49)56:39 88:54
50)64:14 73:13
51)49:32 67:06
52)46:36 69:56
53)63:25 78:52
54)55:49 70:16
55)45:17 77:11
56)67:03 68:53
57)93:40 78:23 22 moves
58)59:35 73:48
59)56:46 77:59
60)59:09 74:54
61)61:21 76:35
62)63:39 77:10
63)50:17 62:07
64)44:48 80:29

Thanks. I agree with you; Komodo is holding too much time in reserve for the endgame, which may never happen. But when we tried to fix this by using the "time use aggressiveness" option test results were not good. Perhaps it's more important to save time for the endgame in these unltra-fast games than in normal ones. Maybe it would test well at a longer tc.

[Ajedrecista]

Hello:

I think that time managament is not the only reason that stockfish is stronger,but I think that komodo played the first moves too fast

Here are the times that the programs had in the clock for the rest of the moves after move 30(first number is stockfish's time).

You can see that in most cases stockfish had less time on the clock after move 30 or in the end of the game if the end happened earlier
(in 54 cases out of 64 cases).

In 19 cases the time gap was more than 20 minutes if I count correctly and in all of them stockfish used more time.


1) 53:42 58:27
2) 61:50 76:59
3) 77:52 89:15 29 moves
4) 88:16 75:41 22 moves
5) 71:11 89:20
6) 38:48 62:20
7) 54:50 51:55
8) 57:55 70:55
9) 53:39 67:21
10)49:21 69:47
11)56:54 81:28
12)49:00 72:38
13)67:27 73:36
14)68:30 62:59
15)45:13 78:56
16)55:18 58:51
17)36:05 71:43
18)45:54 44:57
19)63:18 61:41
20)56:56 78:02
21)62:17 78:16
22)56:04 68:36
23)54:33 71:19
24)39:21 66:01
25)71:04 76:10
26)72:19 78:51
27)41:53 77:28
28)28:23 90:11
29)57:14 73:43
30)55:26 68:33
31)68:54 59:31
32)59:02 69:55
33)53:47 72:30
34)58:03 63:58
35)69:55 74:05
36)60:17 64:44
37)60:05 75:49
38)68:17 80:11
39)32:01 65:17
40)64:02 50:13
41)38:11 55:29
42)65:23 57:39
43)26:33 56:51
44)35:07 64:47
45)68:37 61:27
46)53:21 54:21
47)42:35 44:06
48)39:30 70:06
49)56:39 88:54
50)64:14 73:13
51)49:32 67:06
52)46:36 69:56
53)63:25 78:52
54)55:49 70:16
55)45:17 77:11
56)67:03 68:53
57)93:40 78:23 22 moves
58)59:35 73:48
59)56:46 77:59
60)59:09 74:54
61)61:21 76:35
62)63:39 77:10
63)50:17 62:07
64)44:48 80:29

Thanks. I agree with you; Komodo is holding too much time in reserve for the endgame, which may never happen. But when we tried to fix this by using the "time use aggressiveness" option test results were not good. Perhaps it's more important to save time for the endgame in these unltra-fast games than in normal ones. Maybe it would test well at a longer tc.

Fantastic job Uri! Congratulations.

I computed the time difference in each game and other stuff, just in a try of help:

Code: Select all

Average remaining time for Stockfish                  ~ 3367.5 seconds. 
Sample standard deviation of remaining time for SF    ~  779.4 seconds. 
  
Average remaining time for Komodo                     ~ 4202.6 seconds. 
Sample standard deviation of remaining time for K     ~  607.2 seconds. 
  
Average time difference of (Stockfish - Komodo)       ~ -835.0 seconds. 
Sample standard deviation of time difference (SF - K) ~  854.6 seconds. 
  
======================================================================= 
  
Time difference (SF - K) in each game: 
  
01)  -285 seconds. 
02)  -909 seconds. 
03)  -683 seconds. 
04)   755 seconds. 
05) -1089 seconds. 
06) -1412 seconds. 
07)   175 seconds. 
08)  -780 seconds. 
09)  -822 seconds. 
10) -1226 seconds. 
11) -1474 seconds. 
12) -1418 seconds. 
13)  -369 seconds. 
14)   331 seconds. 
15) -2023 seconds. 
16)  -213 seconds. 
17) -2138 seconds. 
18)    57 seconds. 
19)    97 seconds. 
20) -1266 seconds. 
21)  -959 seconds. 
22)  -752 seconds. 
23) -1006 seconds. 
24) -1600 seconds. 
25)  -306 seconds. 
26)  -392 seconds. 
27) -2135 seconds. 
28) -3708 seconds. 
29)  -989 seconds. 
30)  -787 seconds. 
31)   563 seconds. 
32)  -653 seconds. 
33) -1123 seconds. 
34)  -355 seconds. 
35)  -250 seconds. 
36)  -267 seconds. 
37)  -944 seconds. 
38)  -714 seconds. 
39) -1996 seconds. 
40)   829 seconds. 
41) -1038 seconds. 
42)   464 seconds. 
43) -1818 seconds. 
44) -1780 seconds. 
45)   430 seconds. 
46)   -60 seconds. 
47)   -91 seconds. 
48) -1836 seconds. 
49) -1935 seconds. 
50)  -539 seconds. 
51) -1054 seconds. 
52) -1400 seconds. 
53)  -927 seconds. 
54)  -867 seconds. 
55) -1914 seconds. 
56)  -110 seconds. 
57)   917 seconds. 
58)  -853 seconds. 
59) -1273 seconds. 
60)  -945 seconds. 
61)  -914 seconds. 
62)  -811 seconds. 
63)  -710 seconds. 
64) -2141 seconds. 

Komodo spent less time than Stockfish in 54 games.

I hope no typos.

Regards from Spain.

Ajedrecista.

[Lyudmil Tsvetkov]

No one can beat the mafia.

I am struggling to figure out what that is supposed to signify ??!!

I was thinking something like this:

La Piovra 5 - Il più forte programma di scacchi al mondo. Di pescatori Roberto e Roberto. Con algoritmo di ricerca estensione 'Frutti di Mare'!

Eelco, ma veramente irresistibile.
Siamo tutti Italiani.

There is certain truth about monopolisation, commercial engines suffering, etc. However, without the open source shared code of at least Robbolito and SF most of the good things/impressive developments in the last years would have been impossible.

We would not have had Houdini, we would not have had Stockfish, we would not have had Critter, quite possibly also Komodo. I think this is simply acknowledging the fact that a concerted collective effort is usually more valuable than a single-handed attempt. A shift towards a scientific agenda?