The Distributed Chess Project

Creating Chess-Playing Artificial Neural Networks with Distributed Evolutionary Algorithms

Since 06/23/2002 you have contributed 444 days 9 hours 25 min 27 sec of computing time.

The top 51 performers are (ranking according to test performance, training performance, contributed computing time):

Rank	User-ID	Best Performance on Test Set (%)	Best Performance on Training Set (%)	Cumulated Computing Time (d:h:m:s)
1	esa.elovaara	78	88	38:23:37:41
2	ssieh	76	92	1:2:9:47
3	scotte	76	90	9:23:21:39
4	jason	76	90	4:9:38:23
5	pt.schwarz	76	90	4:4:29:7
6	W	76	86	14:11:55:25
7	gipe	76	86	2:22:34:39
8	m_diosi	74	92	26:8:8:30
9	ad_lord\hotmail.com	74	92	17:12:17:24
10	atalmadge	74	92	7:14:11:42
11	u37206657	74	90	27:16:32:0
12	amiguel	74	90	18:12:28:6
13	david7091	74	90	9:8:34:20
14	5xd7g251lq2vs001	74	90	5:1:56:44
15	gillrich	74	90	3:13:12:34
16	freemac_99	74	88	10:20:14:10
17	njenson	74	88	7:1:47:33
18	jarekfil	74	88	3:21:51:7
19	mitchellonline	74	86	5:0:56:30
20	jean_efpraxiadis	74	84	16:15:9:32
21	rich_luce	74	84	8:20:6:46
22	arnowa	72	94	6:14:56:2
23	ssmythe	72	92	8:12:34:10
24	thunder124	72	92	5:21:18:17
25	toddkloos	72	92	1:18:49:49
26	Sebastian	72	92	1:9:25:17
27	wingy	72	90	13:1:48:49
28	joachim	72	90	10:17:26:20
29	kannan	72	90	7:10:12:14
30	kennynthebest	72	90	6:0:19:37
31	fronte	72	90	3:7:31:4
32	frank	72	90	2:4:59:41
33	sliver	72	90	1:21:28:23
34	aaron	72	90	0:23:47:58
35	midmalex	72	88	20:21:38:25
36	mentis	72	86	20:1:9:18
37	afmomania	72	86	0:3:27:12
38	organizer	72	82	23:20:15:24
39	turrinipa	72	82	6:12:15:0
40	layton97	72	80	3:17:16:44
41	amanwithattitude	70	94	1:23:3:7
42	ic	70	92	13:18:40:8
43	rdmltrs	70	92	9:20:46:55
44	llclarisse	70	92	6:20:58:21
45	arno	70	92	3:10:10:46
46	lshome	70	92	3:3:59:49
47	davidroel	70	90	5:19:51:36
48	paul.abrams	70	90	1:20:1:41
49	joe	70	90	1:13:57:13
50	stjernbecker	70	90	0:22:20:39
51	Barrios_Blue	68	92	6:1:41:49

What I call the 'test set' consists of 50 chess problems with known best continuations. This set must be viewed in contrast to the 'training set' consisting of 50 different problems. The test problems are presented to the net after the training is finished in order to test how well a neural net generalizes from the training problems. This is most important, since a neural net that does well on the training set but not on an independent test set isn't worth its money.

And this is the good news: There definitely is generalization to some degree in the results so far. During training, the neural nets do in fact capture some positional patterns, that enable them to find the correct continuations most of the time when confronted with the test problems.

Numerically, the results in the table are encouraging and rather impressive, because a neural net with e.g. 75% test performance picks the only correct move from a variety of 30 to 50 possible legal moves for 3 out of 4 problems it sees for the very first time. From the pessimist point of view, however, it still fails on 1 out of 4 problems. Moreover, since the typical training performance is higher, we don't have perfect generalization. Therefore it must be the goal to find neural nets with test performance as high as the training performance.

What you as active project participants can and should do at this point is fiddle around with the free parameters of both neural networks and genetic algorithm in order to find a configuration that maximizes the test performance. Any questions? Please check the documentation first, but you may contact me directly as well via Ralf.Seliger@t-online.de