.TL
Results of the AI1 Kalah Tournament
.AB
This report contains the results of the closed AI1
Kalah tournament.  All teams that manage to pass the first stage will
receive bonus points.  The top ten teams receive additional bonus points.
The tournament consists of multiple stages, where agents are disqualified
if they don't perform well enough.
.AE
.NH 1
Stage "Sanity Test"
.LP
All agents are made to compete once against a random bot
on a (6, 6) board.  As the agent is allowed to make the first move, we
know that they must be able to win the game, since this configuration of
Kalah is solved.  To pass this stage, one has to  win a best out of three against
the random bot, otherwise one is disqualified immediately.
.NH 2
Scores
.TS
tab(/) box center;
c | c c c | c
-----
l | n n n | n
.
Agent/Win/Loss/Draw/Score
mo57fery/1/2/0/2
MCTS but in bad/3/0/0/6
Nameless/3/0/0/6
AiAiAiAi/3/0/0/6
Carla, ey, Ei!/3/0/0/6
MyAgent/3/0/0/6
Stirling Archer/3/0/0/6
yahiko/3/0/0/6
ai-1/3/0/0/6
ku81tuli;uj59ames;ez03odak/3/0/0/6
TheBestAgent/3/0/0/6
Agent of Chaos/3/0/0/6
Stickfish/3/0/0/6
na05kymu;ce84pegu/3/0/0/6
jo43beqy/3/0/0/6
EdizDerBreite/3/0/0/6
.TE
.NH 2
Game Log
.TS H
tab(/) box center;
c | c c | c c c
------
n | l l | n n n
.
.TH
Nr./South Agent/North Agent/South/North/Diff.
1/yahiko/Random/57/15/42
2/Nameless/Random/52/20/32
3/TheBestAgent/Random/48/24/24
4/MyAgent/Random/44/28/16
5/TheBestAgent/Random/58/14/44
6/na05kymu;ce84pegu/Random/47/25/22
7/EdizDerBreite/Random/60/12/48
8/ku81tuli;uj59ames;ez03odak/Random/43/29/14
9/Stirling Archer/Random/48/24/24
10/jo43beqy/Random/43/29/14
11/Carla, ey, Ei!/Random/53/19/34
12/Nameless/Random/57/15/42
13/MyAgent/Random/59/13/46
14/Stirling Archer/Random/53/19/34
15/Nameless/Random/57/15/42
16/Stirling Archer/Random/49/23/26
17/Stickfish/Random/52/20/32
18/Agent of Chaos/Random/43/29/14
19/yahiko/Random/58/14/44
20/EdizDerBreite/Random/48/24/24
21/MCTS but in bad/Random/55/17/38
22/na05kymu;ce84pegu/Random/51/21/30
23/jo43beqy/Random/44/28/16
24/ai-1/Random/49/23/26
25/jo43beqy/Random/42/30/12
26/Carla, ey, Ei!/Random/59/13/46
27/ku81tuli;uj59ames;ez03odak/Random/62/10/52
28/mo57fery/Random/38/34/4
29/mo57fery/Random/34/38/-4
30/Carla, ey, Ei!/Random/48/24/24
31/na05kymu;ce84pegu/Random/49/23/26
32/TheBestAgent/Random/57/15/42
33/Agent of Chaos/Random/53/19/34
34/MCTS but in bad/Random/54/18/36
35/mo57fery/Random/30/42/-12
36/MyAgent/Random/45/27/18
37/ku81tuli;uj59ames;ez03odak/Random/60/12/48
38/yahiko/Random/43/29/14
39/Agent of Chaos/Random/50/22/28
40/AiAiAiAi/Random/56/16/40
41/Stickfish/Random/61/11/50
42/EdizDerBreite/Random/59/13/46
43/AiAiAiAi/Random/61/11/50
44/AiAiAiAi/Random/43/29/14
45/MCTS but in bad/Random/46/26/20
46/Stickfish/Random/50/22/28
47/ai-1/Random/47/25/22
48/ai-1/Random/41/31/10
.TE
.PP
These agents were disqualified
for failing to meet the necessary criteria for proceeding to the
next round:
.IP \(bu
mo57fery
.NH 1
Stage "Round Robin (6, 6)"
.LP
All agents play against all other agents, on both sides
of a Kalah board.  If one agent definitively manages to beat another agent,
they are awarded two points, and the opponent is given no points.
For a draw, both agents are granted a single point.  The final score of this
round is calculated by summing up the points for each game.  Agents that suffered
more losses than wins are disqualifed.
.NH 2
Scores
.TS
tab(/) box center;
c | c c c | c
-----
l | n n n | n
.
Agent/Win/Loss/Draw/Score
jo43beqy/0/28/0/0
Agent of Chaos/5/23/0/10
ai-1/5/21/2/12
yahiko/7/19/1/15
TheBestAgent/7/20/1/15
AiAiAiAi/8/20/0/16
Nameless/12/14/0/24
MyAgent/12/13/1/25
na05kymu;ce84pegu/14/12/0/28
ku81tuli;uj59ames;ez03odak/14/12/1/29
Stirling Archer/16/11/0/32
EdizDerBreite/23/4/0/46
MCTS but in bad/23/4/1/47
Stickfish/23/2/1/47
Carla, ey, Ei!/25/3/0/50
.TE
.NH 2
Game Log
.TS H
tab(/) box center;
c | c c | c c c
------
n | l l | n n n
.
.TH
Nr./South Agent/North Agent/South/North/Diff.
1/jo43beqy/MyAgent/23/49/-26
2/MCTS but in bad/Agent of Chaos/51/21/30
3/Agent of Chaos/yahiko/29/43/-14
4/yahiko/na05kymu;ce84pegu/29/43/-14
5/Nameless/ai-1/54/18/36
6/Agent of Chaos/EdizDerBreite/20/52/-32
7/MyAgent/yahiko/33/39/-6
8/Agent of Chaos/AiAiAiAi/42/30/12
9/MCTS but in bad/EdizDerBreite/41/31/10
10/TheBestAgent/Stickfish/27/45/-18
11/MCTS but in bad/Stickfish/24/48/-24
12/MCTS but in bad/na05kymu;ce84pegu/42/30/12
13/Carla, ey, Ei!/ai-1/54/18/36
14/Stirling Archer/MyAgent/43/29/14
15/na05kymu;ce84pegu/AiAiAiAi/44/28/16
16/MyAgent/Stirling Archer/21/34/-13
17/ai-1/MCTS but in bad/17/55/-38
18/ku81tuli;uj59ames;ez03odak/AiAiAiAi/33/39/-6
19/Stickfish/MyAgent/41/31/10
20/EdizDerBreite/ku81tuli;uj59ames;ez03odak/46/26/20
21/TheBestAgent/Agent of Chaos/43/29/14
22/ai-1/Stickfish/33/39/-6
23/Stickfish/jo43beqy/49/23/26
24/ku81tuli;uj59ames;ez03odak/MCTS but in bad/29/43/-14
25/MyAgent/na05kymu;ce84pegu/17/35/-18
26/Stickfish/Stirling Archer/41/31/10
27/na05kymu;ce84pegu/MCTS but in bad/31/41/-10
28/jo43beqy/na05kymu;ce84pegu/18/54/-36
29/MCTS but in bad/MyAgent/37/35/2
30/Agent of Chaos/Stickfish/27/45/-18
31/jo43beqy/TheBestAgent/24/48/-24
32/yahiko/jo43beqy/52/20/32
33/Nameless/jo43beqy/50/22/28
34/ai-1/Stirling Archer/19/53/-34
35/yahiko/Carla, ey, Ei!/18/54/-36
36/yahiko/MyAgent/20/52/-32
37/jo43beqy/Nameless/22/50/-28
38/na05kymu;ce84pegu/Carla, ey, Ei!/32/40/-8
39/EdizDerBreite/yahiko/58/14/44
40/Agent of Chaos/TheBestAgent/34/38/-4
41/na05kymu;ce84pegu/TheBestAgent/37/35/2
42/Agent of Chaos/Nameless/17/55/-38
43/Stickfish/Nameless/47/25/22
44/na05kymu;ce84pegu/MyAgent/42/30/12
45/Nameless/AiAiAiAi/49/23/26
46/jo43beqy/Agent of Chaos/32/40/-8
47/TheBestAgent/Carla, ey, Ei!/21/51/-30
48/Nameless/Stirling Archer/19/53/-34
49/na05kymu;ce84pegu/Stickfish/34/38/-4
50/ku81tuli;uj59ames;ez03odak/TheBestAgent/47/25/22
51/Agent of Chaos/Stirling Archer/22/50/-28
52/na05kymu;ce84pegu/Agent of Chaos/45/27/18
53/ai-1/na05kymu;ce84pegu/24/48/-24
54/yahiko/Agent of Chaos/43/29/14
55/MCTS but in bad/jo43beqy/52/20/32
56/ku81tuli;uj59ames;ez03odak/Stickfish/28/44/-16
57/ai-1/MyAgent/14/58/-44
58/yahiko/MCTS but in bad/20/52/-32
59/Stickfish/na05kymu;ce84pegu/45/27/18
60/Agent of Chaos/ku81tuli;uj59ames;ez03odak/30/42/-12
61/MCTS but in bad/ai-1/46/26/20
62/ku81tuli;uj59ames;ez03odak/Carla, ey, Ei!/34/38/-4
63/AiAiAiAi/Nameless/19/53/-34
64/EdizDerBreite/Stickfish/34/23/11
65/MCTS but in bad/Carla, ey, Ei!/33/39/-6
66/ku81tuli;uj59ames;ez03odak/jo43beqy/45/27/18
67/yahiko/ai-1/31/41/-10
68/Carla, ey, Ei!/jo43beqy/58/14/44
69/MyAgent/ku81tuli;uj59ames;ez03odak/36/36/0
70/ku81tuli;uj59ames;ez03odak/Stirling Archer/19/35/-16
71/TheBestAgent/MCTS but in bad/20/52/-32
72/TheBestAgent/ai-1/19/53/-34
73/EdizDerBreite/MCTS but in bad/49/23/26
74/jo43beqy/Carla, ey, Ei!/24/48/-24
75/Agent of Chaos/MCTS but in bad/23/49/-26
76/Agent of Chaos/Carla, ey, Ei!/21/51/-30
77/ai-1/yahiko/36/36/0
78/AiAiAiAi/Carla, ey, Ei!/29/43/-14
79/jo43beqy/Stirling Archer/10/62/-52
80/ku81tuli;uj59ames;ez03odak/MyAgent/37/35/2
81/jo43beqy/EdizDerBreite/13/59/-46
82/jo43beqy/yahiko/21/51/-30
83/MyAgent/jo43beqy/46/26/20
84/yahiko/EdizDerBreite/20/52/-32
85/MyAgent/Agent of Chaos/54/18/36
86/AiAiAiAi/EdizDerBreite/27/45/-18
87/ku81tuli;uj59ames;ez03odak/na05kymu;ce84pegu/45/27/18
88/EdizDerBreite/Agent of Chaos/48/24/24
89/Stickfish/yahiko/43/29/14
90/Carla, ey, Ei!/TheBestAgent/44/28/16
91/Stickfish/MCTS but in bad/36/36/0
92/EdizDerBreite/Nameless/47/25/22
93/na05kymu;ce84pegu/Stirling Archer/27/31/-4
94/Carla, ey, Ei!/AiAiAiAi/40/32/8
95/jo43beqy/Stickfish/32/40/-8
96/MyAgent/Nameless/24/48/-24
97/Carla, ey, Ei!/na05kymu;ce84pegu/44/28/16
98/MCTS but in bad/AiAiAiAi/47/25/22
99/Stirling Archer/AiAiAiAi/45/27/18
100/Stickfish/ai-1/55/17/38
101/ai-1/Nameless/21/51/-30
102/TheBestAgent/MyAgent/28/44/-16
103/AiAiAiAi/ku81tuli;uj59ames;ez03odak/31/41/-10
104/ku81tuli;uj59ames;ez03odak/ai-1/51/21/30
105/na05kymu;ce84pegu/jo43beqy/46/26/20
106/Nameless/Carla, ey, Ei!/32/40/-8
107/TheBestAgent/Nameless/38/34/4
108/jo43beqy/AiAiAiAi/30/42/-12
109/Nameless/MCTS but in bad/27/45/-18
110/Agent of Chaos/MyAgent/27/45/-18
111/Nameless/Stickfish/27/45/-18
112/ai-1/AiAiAiAi/29/43/-14
113/AiAiAiAi/Stickfish/30/42/-12
114/Stickfish/Carla, ey, Ei!/35/37/-2
115/yahiko/Stickfish/19/53/-34
116/MyAgent/Carla, ey, Ei!/28/44/-16
117/Stirling Archer/Carla, ey, Ei!/31/41/-10
118/Carla, ey, Ei!/Stirling Archer/42/30/12
119/MyAgent/AiAiAiAi/48/24/24
120/AiAiAiAi/Agent of Chaos/57/15/42
121/MyAgent/TheBestAgent/51/21/30
122/EdizDerBreite/ai-1/51/21/30
123/ai-1/TheBestAgent/36/36/0
124/Nameless/TheBestAgent/34/23/11
125/Stickfish/EdizDerBreite/24/48/-24
126/yahiko/AiAiAiAi/32/40/-8
127/yahiko/Stirling Archer/20/52/-32
128/Stirling Archer/EdizDerBreite/26/30/-4
129/EdizDerBreite/AiAiAiAi/40/32/8
130/Stirling Archer/TheBestAgent/45/27/18
131/ai-1/EdizDerBreite/11/61/-50
132/AiAiAiAi/jo43beqy/50/22/28
133/Nameless/ku81tuli;uj59ames;ez03odak/37/35/2
134/TheBestAgent/AiAiAiAi/37/35/2
135/Stickfish/TheBestAgent/39/33/6
136/Stirling Archer/yahiko/44/28/16
137/ai-1/Carla, ey, Ei!/22/50/-28
138/MyAgent/ai-1/45/27/18
139/MCTS but in bad/yahiko/51/21/30
140/na05kymu;ce84pegu/Nameless/42/30/12
141/Nameless/MyAgent/34/38/-4
142/Agent of Chaos/jo43beqy/50/22/28
143/Carla, ey, Ei!/yahiko/50/22/28
144/Stirling Archer/ai-1/44/28/16
145/ku81tuli;uj59ames;ez03odak/Nameless/42/30/12
146/Nameless/Agent of Chaos/44/28/16
147/jo43beqy/MCTS but in bad/19/53/-34
148/ku81tuli;uj59ames;ez03odak/EdizDerBreite/26/46/-20
149/yahiko/Nameless/15/57/-42
150/Stickfish/Agent of Chaos/44/28/16
151/na05kymu;ce84pegu/ku81tuli;uj59ames;ez03odak/33/39/-6
152/TheBestAgent/EdizDerBreite/31/41/-10
153/Nameless/yahiko/48/24/24
154/Carla, ey, Ei!/ku81tuli;uj59ames;ez03odak/46/26/20
155/Nameless/EdizDerBreite/27/28/-1
156/MCTS but in bad/Stirling Archer/37/35/2
157/MyAgent/Stickfish/31/41/-10
158/ai-1/ku81tuli;uj59ames;ez03odak/35/37/-2
159/Stirling Archer/jo43beqy/41/31/10
160/TheBestAgent/ku81tuli;uj59ames;ez03odak/28/44/-16
161/MCTS but in bad/Nameless/41/31/10
162/MCTS but in bad/TheBestAgent/47/25/22
163/yahiko/TheBestAgent/39/33/6
164/Stirling Archer/ku81tuli;uj59ames;ez03odak/42/30/12
165/MCTS but in bad/ku81tuli;uj59ames;ez03odak/42/30/12
166/TheBestAgent/na05kymu;ce84pegu/29/43/-14
167/EdizDerBreite/TheBestAgent/40/32/8
168/Agent of Chaos/na05kymu;ce84pegu/30/42/-12
169/AiAiAiAi/yahiko/58/14/44
170/jo43beqy/ku81tuli;uj59ames;ez03odak/22/50/-28
171/Stirling Archer/Stickfish/36/22/14
172/Stirling Archer/Agent of Chaos/40/32/8
173/AiAiAiAi/MCTS but in bad/27/45/-18
174/TheBestAgent/jo43beqy/39/33/6
175/Carla, ey, Ei!/Agent of Chaos/54/18/36
176/MyAgent/EdizDerBreite/31/41/-10
177/na05kymu;ce84pegu/yahiko/47/25/22
178/Stirling Archer/MCTS but in bad/29/43/-14
179/AiAiAiAi/TheBestAgent/42/30/12
180/MyAgent/MCTS but in bad/35/37/-2
181/Carla, ey, Ei!/EdizDerBreite/35/37/-2
182/Carla, ey, Ei!/MCTS but in bad/45/27/18
183/na05kymu;ce84pegu/EdizDerBreite/21/51/-30
184/TheBestAgent/Stirling Archer/40/32/8
185/EdizDerBreite/Carla, ey, Ei!/42/30/12
186/yahiko/ku81tuli;uj59ames;ez03odak/40/32/8
187/AiAiAiAi/na05kymu;ce84pegu/31/25/6
188/Stickfish/ku81tuli;uj59ames;ez03odak/41/31/10
189/AiAiAiAi/MyAgent/13/59/-46
190/Stirling Archer/na05kymu;ce84pegu/40/32/8
191/AiAiAiAi/Stirling Archer/26/46/-20
192/EdizDerBreite/MyAgent/16/13/3
193/TheBestAgent/yahiko/32/29/3
194/jo43beqy/ai-1/27/45/-18
195/AiAiAiAi/ai-1/23/49/-26
196/Agent of Chaos/ai-1/38/34/4
197/ai-1/Agent of Chaos/33/39/-6
198/Carla, ey, Ei!/MyAgent/40/32/8
199/EdizDerBreite/na05kymu;ce84pegu/45/27/18
200/EdizDerBreite/jo43beqy/52/20/32
201/na05kymu;ce84pegu/ai-1/49/23/26
202/ku81tuli;uj59ames;ez03odak/Agent of Chaos/43/29/14
203/Nameless/na05kymu;ce84pegu/33/39/-6
204/EdizDerBreite/Stirling Archer/41/31/10
205/ku81tuli;uj59ames;ez03odak/yahiko/50/22/28
206/Carla, ey, Ei!/Nameless/47/25/22
207/Stickfish/AiAiAiAi/48/24/24
208/ai-1/jo43beqy/43/29/14
209/Stirling Archer/Nameless/50/22/28
210/Carla, ey, Ei!/Stickfish/33/39/-6
.TE
.PP
These agents were disqualified
for failing to meet the necessary criteria for proceeding to the
next round:
.IP \(bu
Nameless
.IP \(bu
AiAiAiAi
.IP \(bu
yahiko
.IP \(bu
ai-1
.IP \(bu
TheBestAgent
.IP \(bu
Agent of Chaos
.IP \(bu
jo43beqy
.NH 1
Stage "Round Robin (8, 8)"
.LP
All agents play against all other agents, on both sides
of a Kalah board.  If one agent definitively manages to beat another agent,
they are awarded two points, and the opponent is given no points.
For a draw, both agents are granted a single point.  The final score of this
round is calculated by summing up the points for each game.  Agents that suffered
more losses than wins are disqualifed.
.NH 2
Scores
.TS
tab(/) box center;
c | c c c | c
-----
l | n n n | n
.
Agent/Win/Loss/Draw/Score
MyAgent/1/12/0/2
ku81tuli;uj59ames;ez03odak/1/11/1/3
na05kymu;ce84pegu/4/9/0/8
MCTS but in bad/6/8/0/12
Carla, ey, Ei!/7/5/1/15
Stirling Archer/8/6/0/16
Stickfish/11/3/0/22
EdizDerBreite/13/1/0/26
.TE
.NH 2
Game Log
.TS H
tab(/) box center;
c | c c | c c c
------
n | l l | n n n
.
.TH
Nr./South Agent/North Agent/South/North/Diff.
1/MyAgent/EdizDerBreite/42/86/-44
2/Stickfish/Stirling Archer/54/74/-20
3/na05kymu;ce84pegu/MyAgent/76/52/24
4/Stirling Archer/ku81tuli;uj59ames;ez03odak/73/55/18
5/Carla, ey, Ei!/EdizDerBreite/51/77/-26
6/EdizDerBreite/MCTS but in bad/90/38/52
7/na05kymu;ce84pegu/ku81tuli;uj59ames;ez03odak/80/48/32
8/EdizDerBreite/Carla, ey, Ei!/68/60/8
9/Stickfish/EdizDerBreite/74/54/20
10/MyAgent/na05kymu;ce84pegu/63/65/-2
11/na05kymu;ce84pegu/EdizDerBreite/60/68/-8
12/MyAgent/Stirling Archer/57/60/-3
13/MCTS but in bad/MyAgent/76/52/24
14/Carla, ey, Ei!/na05kymu;ce84pegu/78/50/28
15/Stirling Archer/EdizDerBreite/55/73/-18
16/Stickfish/Carla, ey, Ei!/74/54/20
17/Stickfish/MyAgent/72/56/16
18/Stickfish/ku81tuli;uj59ames;ez03odak/84/44/40
19/ku81tuli;uj59ames;ez03odak/MyAgent/73/55/18
20/ku81tuli;uj59ames;ez03odak/na05kymu;ce84pegu/59/63/-4
21/ku81tuli;uj59ames;ez03odak/EdizDerBreite/26/102/-76
22/Stirling Archer/Carla, ey, Ei!/73/55/18
23/ku81tuli;uj59ames;ez03odak/MCTS but in bad/62/66/-4
24/MCTS but in bad/Stirling Archer/51/77/-26
25/MCTS but in bad/EdizDerBreite/56/72/-16
26/na05kymu;ce84pegu/MCTS but in bad/60/68/-8
27/Stickfish/na05kymu;ce84pegu/89/39/50
28/MyAgent/MCTS but in bad/51/77/-26
29/Stirling Archer/na05kymu;ce84pegu/64/50/14
30/MCTS but in bad/Stickfish/55/73/-18
31/ku81tuli;uj59ames;ez03odak/Carla, ey, Ei!/64/64/0
32/na05kymu;ce84pegu/Stirling Archer/68/60/8
33/Carla, ey, Ei!/Stirling Archer/90/38/52
34/MyAgent/Carla, ey, Ei!/55/73/-18
35/na05kymu;ce84pegu/Stickfish/52/76/-24
36/Carla, ey, Ei!/MyAgent/76/52/24
37/MCTS but in bad/ku81tuli;uj59ames;ez03odak/76/52/24
38/Stirling Archer/MCTS but in bad/73/55/18
39/MyAgent/Stickfish/60/68/-8
40/EdizDerBreite/Stirling Archer/86/42/44
41/MCTS but in bad/na05kymu;ce84pegu/74/54/20
42/na05kymu;ce84pegu/Carla, ey, Ei!/63/48/15
43/Carla, ey, Ei!/Stickfish/60/68/-8
44/EdizDerBreite/MyAgent/90/38/52
45/MyAgent/ku81tuli;uj59ames;ez03odak/84/44/40
46/Carla, ey, Ei!/MCTS but in bad/80/48/32
47/Stirling Archer/Stickfish/68/60/8
48/EdizDerBreite/na05kymu;ce84pegu/90/38/52
49/EdizDerBreite/Stickfish/73/55/18
50/EdizDerBreite/ku81tuli;uj59ames;ez03odak/82/46/36
51/Stickfish/MCTS but in bad/80/48/32
52/MCTS but in bad/Carla, ey, Ei!/56/72/-16
53/ku81tuli;uj59ames;ez03odak/Stickfish/53/75/-22
54/Stirling Archer/MyAgent/76/52/24
55/Carla, ey, Ei!/ku81tuli;uj59ames;ez03odak/74/54/20
56/ku81tuli;uj59ames;ez03odak/Stirling Archer/61/67/-6
.TE
.PP
These agents were disqualified
for failing to meet the necessary criteria for proceeding to the
next round:
.IP \(bu
MyAgent
.IP \(bu
na05kymu;ce84pegu
.IP \(bu
ku81tuli;uj59ames;ez03odak
.IP \(bu
MCTS but in bad
.NH 1
Stage "Round Robin (10, 10)"
.LP
All agents play against all other agents, on both sides
of a Kalah board.  If one agent definitively manages to beat another agent,
they are awarded two points, and the opponent is given no points.
For a draw, both agents are granted a single point.  The final score of this
round is calculated by summing up the points for each game.  Agents that suffered
more losses than wins are disqualifed.
.NH 2
Scores
.TS
tab(/) box center;
c | c c c | c
-----
l | n n n | n
.
Agent/Win/Loss/Draw/Score
Carla, ey, Ei!/1/5/0/2
Stirling Archer/1/5/0/2
Stickfish/5/1/0/10
EdizDerBreite/5/1/0/10
.TE
.NH 2
Game Log
.TS H
tab(/) box center;
c | c c | c c c
------
n | l l | n n n
.
.TH
Nr./South Agent/North Agent/South/North/Diff.
1/Carla, ey, Ei!/Stickfish/80/120/-40
2/Stirling Archer/EdizDerBreite/87/113/-26
3/Stickfish/Carla, ey, Ei!/107/93/14
4/EdizDerBreite/Stirling Archer/117/83/34
5/EdizDerBreite/Carla, ey, Ei!/123/77/46
6/Stickfish/EdizDerBreite/89/111/-22
7/Carla, ey, Ei!/EdizDerBreite/92/108/-16
8/Carla, ey, Ei!/Stirling Archer/115/85/30
9/Stirling Archer/Carla, ey, Ei!/121/79/42
10/Stirling Archer/Stickfish/87/113/-26
11/Stickfish/Stirling Archer/116/84/32
12/EdizDerBreite/Stickfish/96/104/-8
.TE
.PP
These agents were disqualified
for failing to meet the necessary criteria for proceeding to the
next round:
.IP \(bu
Carla, ey, Ei!
.IP \(bu
Stirling Archer
.NH 1
Stage "Round Robin (12, 12)"
.LP
All agents play against all other agents, on both sides
of a Kalah board.  If one agent definitively manages to beat another agent,
they are awarded two points, and the opponent is given no points.
For a draw, both agents are granted a single point.  The final score of this
round is calculated by summing up the points for each game.  Agents that suffered
more losses than wins are disqualifed.
.NH 2
Scores
.TS
tab(/) box center;
c | c c c | c
-----
l | n n n | n
.
Agent/Win/Loss/Draw/Score
Stickfish/0/2/0/0
EdizDerBreite/2/0/0/4
.TE
.NH 2
Game Log
.TS H
tab(/) box center;
c | c c | c c c
------
n | l l | n n n
.
.TH
Nr./South Agent/North Agent/South/North/Diff.
1/Stickfish/EdizDerBreite/135/153/-18
2/EdizDerBreite/Stickfish/149/139/10
.TE
.PP
These agents were disqualified
for failing to meet the necessary criteria for proceeding to the
next round:
.IP \(bu
Stickfish
.NH 1
Final score
.LP
The top ten agents are as follows:
.IP 1
EdizDerBreite (Score: 92)
.IP 2
Stickfish (Score: 85)
.IP 3
Carla, ey, Ei! (Score: 73)
.IP 4
MCTS but in bad (Score: 65)
.IP 5
Stirling Archer (Score: 56)
.IP 6
na05kymu;ce84pegu (Score: 42)
.IP 7
ku81tuli;uj59ames;ez03odak (Score: 38)
.IP 8
MyAgent (Score: 33)
.IP 9
Nameless (Score: 30)
.IP 10
AiAiAiAi (Score: 22)
.LP
Congratulations to all participating teams!
.PP
The remaining scores are:
.IP 11
yahiko (Score: 21)
.IP 11
TheBestAgent (Score: 21)
.IP 12
ai-1 (Score: 18)
.IP 13
Agent of Chaos (Score: 16)
.IP 14
jo43beqy (Score: 6)
.IP 15
mo57fery (Score: 2)

.bp
.NH 1
About the agents

.LP
This section contains the abridged, partly spellchecked and expanded (where
.I needed )
contents of the
.CW ABOUT
file, if it was submitted.
The agent names are taken from the submitted code where it was possible.
Else the name consists of the idm ids of the submitters.

.NH 2
mo57fery (224972)
.SH
.LP
Make the game a minmax problem and choose the best value to continue to play, without any assumption that the other makes a mistake.
.SH
.LP
The minmax function calls itself recursively. It searches the best moves and the best value.

.NH 2
MCTS but in bad (223785)
.SH
Idea
.LP
Monte Carlo tree search
.SH
Implementation
.LP
Concepts used:
.IP \(bu
MCTS with board heuristic instead of playouts
.IP \(bu
UCT-Score for search guidance (
.CW c
= 2.5)
.IP \(bu
Min-Max Backup
  
The ideas came from the paper ``Trade-Offs in Sampling-Based Adversarial Planning``.

Saving the built tree to avoid recomputation was attempted, but the overhead created by the multiprocessing manager made in inpractical.

.NH 2
Nameless (224092)
.SH
Idea
.LP
The idea is to use MCTS to search the game tree via biased sampling and then take the move, which seems to lead to best mean outcomes.
.SH
Implementation
.LP
Our agent was written as follows:
A combination of Monte-Carlo-Tree-Search and Min-Max with Alpha-Beta-Pruning was used.
The agent starts with mcts and switches at a specific state of the game to min-max.
The turning point is chosen by the number of seeds left in the game. Mcts should perform better
at the beginning of the game since the branching factor is probably too high for min-max to give good results.
Mcts develops a good intuition of the game and min-max finishes the game better.

The tree for mcts is build out of Node objects, which save the board state, the q value, the number of visits, their parent node,
their explored children and the unexplored legal moves. With two convenience functions, one to add a rollout result and
one to extend the tree by a node. The remaining agent consists of an uct_score function to calculate the score used
to select the best child, a select function which uses the uct score to select the best child, a rollout function, which
follows a random policy until the game terminates, a backprop_result function which backpropagates the result of the rollout
up the tree, an evaluate function, which calculate the differences in stores between both players and a q function, which decides
base on this difference, whether its a win, a tie or a loss.
The one level above is the mcts_search function, which either extends a node, starts a rollout and then backpropagates the result or
selects the best child, if all the nodes children are already extended. This is done in a time restricted loop and once
the time is up the action with the best mean outcome will be returned.
At the highest level the agent generator (which yields the mcts_search) is called with the actual board state as
.CW root_node .
This can be easily improved by taking the child of the previous root corresponding to the done action. This enables to reuse some of the previous search effort, but also takes up more RAM.

The min-max search given was adapted with an evaluation which counts the number of seeds a player has captured minus the number of 
seeds an opponent has captured and also takes into account the difference of seeds still in the pits between the opponent and us. Furthermore alpha-beta pruning was added to the min-max algorithm with variable ordering by evaluating the next board state.

.NH 2
AiAiAiAi (224180)
.SH
Idea
.LP
Basic min/max algorithm with alpha/beta pruning but without more sophisticated methods.
.SH
Implementation
.LP
Simple evaluation function that takes into account the tokens in the stores with a factor for scaling, simple minimax algorithm like presented in the lecture.
The agent was written in Python.

.NH 2
Carla, ey, Ei! (224221)
.SH
Idea
.LP
We started in Python, then switched to Golang because of the lack of speed in Python.
First we implemented a simple MinMax algorithm with alpha-beta pruning.
To deliver multiple outputs in the given time, we included iterative deepening.
As we read more about the game theory of Kalah, we turned the MinMax algorithm into a NegaMax algorithm, because Kalah is a zero-sum game.
After further reading we implemented a simple move order to speed up the search by using more effective pruning.
In Python we also tested an implementation of a transposition table that stores game states and their best moves.
This was difficult because kgp.py used processes rather than threads, so the table was not saved between searches.
So we switched to principal variation search, which assumes that the first move in the order is the best, and is even more effective at pruning.
But in the worst case it is as fast as NegaMax.
This is where Python reached its limits and we switched to a language that is compiled.
Go was a good option because it had a really good tutorial and an already implemented Kalah game protocol.
.SH
Implementation
.LP
We do iterative deepening on the
.CW Search
function
There we use the principal variation search based on the pseudocode on wikipedia with the following parameters:
.IP \(bu
.CW state
(the game state of the node)
.IP \(bu
.CW side
(the player whose turn it is)
.IP \(bu
.CW depth
(the search depth)
.IP \(bu
.CW alpha
(value for pruning)
.IP \(bu
.CW beta
(value for pruning)

.LP
For the evaluation we chose a rather simple heuristic which is just the score difference.

Now comes the main part of the principal variation search:
Inside the for loop we traverse all possible moves for the given state.
Before searching further nodes we check if the enemy has already won, and if so we continue with the next move.
First we evaluate the first move we got from ordering. Here we search the child nodes recursively.
If the move grants us another turn, we leave the alpha and beta bounds as they are.
Otherwise we search as in NegaMax with alpha = -beta and beta = -alpha. The score is also negated.
Next we search all other moves in the order, but with a smaller search window for alpha = -alpha-1 and beta = -alpha.
If the search fails high, meaning that the calculated score is between the initial alpha and beta, we have to search the nodes with the boundaries again as we searched the first move.

Before searching the next move in the for loop we compare the alpha and the calculated score. If the new score is better, we update alpha and bestMove. And here we prune if the alpha is greater than or equal to the beta.

Finally we return
.CW alpha
and the
.CW bestMove .

.NH 2
MyAgent (224225)
.SH
Idea
.LP
Minimax-search with iterative deepening search
.SH
Implementation
.LP
We implemented a negamax algorithm which is basically the minimax-algorithm where min and max use the same function just with different signs.
We also tried to implement alpha-beta pruning but couldn't get that working.
Thus we are submitting the agent without pruning. We programmed the agent in java using recursion.

.NH 2
Stirling Archer (224474)
.SH
Idea
.LP
The agent performs a iterative deepening search applying multiple performance enhancements.
First off alpha beta pruning is used to reduce the expanded part of the search tree.
To further improve the effectiveness of the pruning move ordering is implemented.
We try to predict the potential for each allowed move by looking at move properties that can be directly observed:
Properties seen as good are (in descending order):

.IP 1.
the player is allowed to take another move directly after this move (this is essentially a ``free`` move)
.IP 2.
the move captures a pit (thus ``steeling`` seeds from the enemy side and securing multiple seeds in one move)
.IP 3.
the move added seeds to the players store (going one or multiple times over the own store)

.LP
Moves are expanded in this order.
Moves that do not have any of these properties are expanded last.

Our evaluation function takes three key measures into account:

First: the difference between the own store and the enemy store.
Obviously this is a clear indicator if one side is closer to winning.

Second: the number of seeds on the enemy side.
Seeds on the enemy side are potentially easier to get into the enemy store.
This is not as strong an indicator as the seeds that are already in the stores an thus weighted lower.

Third: if one player already won, all other metrics should be irrelevant and a huge constant is added.
This ranks winning moves always above non-winning ones but also allows to distinguish between ``better`` and ``worse`` wins.
This is relevant in the case that it an enemy move was made on this branch.
We assumed the enemy to take the best move possible according to
.B our
evaluation function.
Since the enemy might have a different evaluation function it might take a different path than we assumed that might have truncated.
Using a combination of the non-winning-evaluation and the winning constant allows us the estimate to overall-goodness of the direction of a (potentially winning) move.

.SH
Implementation
.LP
Our agent is implemented in python and uses the provided template.
For the iterative deepening we start with depth one.
This depth should be reached under all hardware conditions and bord sizes.
After that we continue with depth two and increase the depth after each successful run by two.

The calculation of the best move for each depth is done in
.CW search_alpha_beta
which calls itself recursively.
For each recursive call we take into account, whether the current player is allowed to make another move or not.
If the player is is allowed to make another move we calculate the next best move by calling
.CW search_alpha_beta
again with the same parameter (but progressed bord state).
If the player is not allowed to make another move
.CW search_alpha_beta
is called with switched player perspective and decreased depth.
So the depth corresponds to the number times the player switch turns and not to the moves that are made.
We also implemented a mechanism to detect whether a branch led to the win of any player.
In this case we stop the recursion so no time is wasted on this branch.

In the calculation of evaluation function uses two magic constants:

First: If one player won, we return a large constant of 1000000 if we won or -1000000 if we lost.
This constants should be large enough to exceed every evaluation value generated in a non-final case.

Second: The seeds on the enemy side are counted and multiplied by 0.1 and then subtracted from the evaluation.
This part of the evaluation should be seen as a tie breaker for two moves that otherwise would yield (nearly) identical differences in stored seeds.
As a tie breaker it should not influence the evaluation function too strongly.
The factor 0.1 was empirically proven to be useful.

.NH 2
yahiko (224480)
.SH
Idea
.LP
Simple minmax and complex evaluation function based on game stage.
More importance to capture moves at later stages of game.
Game stage decided based on number of seeds in house.
.SH
.LP
Minmax algorithm with complex evaluation function.
The agent was written in Java.

.NH 2
ai-1 (224535)
.SH
Idea
.LP
Minimax Search algorithm with Alpha-Beta Pruning
.SH
Implementation
.LP
.IP 1.
Run Minimax Alpha-Beta Pruning with increasing depth (2, 3, 5, 8, ...)
.IP 2.
The moves are in descending order with respect to the evaluation score
.IP 3.
Evaluation score is a linear combination of:
.QS
.IP \(bu
Store Advantage
.IP \(bu
Stone Differences
.IP \(bu
Capture Potential and Loss Potential
.IP \(bu
Extra bonus if agent can move again
.QE

.NH 2
ku81tuli;uj59ames;ez03odak (224684)
.SH
Idea
.LP
While having a minimax search with alpha-beta pruning and iterative deepening search, the main effort has been put into implementing a suitable evaluation function, like suggested in the paper.
Simple move ordering techniques like searching for bonus moves first have also been tried out, but the algorithm has not shown significant improvements compared to this basic version.
Therefore, we decided to submit the original version without move ordering.

.SH
Implementation
.LP
The agent has been implemented in python using the provided
.CW kgp.py
file for the connection.

.NH 2
TheBestAgent (224715)
.SH
Idea
.LP
The agent uses the minmax algorithm with a custom evaluation function.

The evaluation function calculates the utility based on the number of seeds on each side of the board.
It adds a quarter of the total seeds on the
.CW SOUTH
side and the number of seeds in the
.CW SOUTH
store to the utility.
It then subtracts a quarter of the total seeds on the
.CW NORTH
side and the number of seeds in the
.CW NORTH
store from the utility.

The use of 0.25 as a multiplier makes sure that the AI values having seeds in its own side of the board, but not as much as having seeds in its store.
This is is a ``long term`` protection to prevent having too few seeds on the maximizing side and loosing due to no seeds left on the side.
.SH
Implementation
.LP
The agent was written in Python.

.NH 2
Agent of Chaos (224732)
.SH
Idea
.LP
We used the minmax algorithm with alpha-beta pruning and move ordering.
.SH
Implementation
.LP
The agent was written in Java.

.NH 2
Stickfish (224734)
.SH
Idea
.LP
Alpha Beta Pruning with iterative deepening

.SH
Move ordering
.LP
.IP \(bu
Extra moves from right to left
.IP \(bu
Steals from left to right, but only if there are no bonus moves
.IP \(bu
Other moves from right to left

.LP
Apart from that we change the moveorder depending on the results of our searches by moving the best move to the front
.SH
Evaluation
.IP \(bu
Simulate all possible extra moves from right to left (go down only one branch of extra moves)  
.IP \(bu
Check the steal granting the most points.
We subtract -0.75 to the side currently moving because we already simulated so many moves for it  
if the sum of the seeds not yet in a store are smaller than 3 *
.CW board_size
we give a bonus to the side with more seeds on their half (because they keep them if the game ends)
Do not ask how we came up with the values for these things (hint: it was not a genetic algorithm)
.IP \(bu
Simulating the moves makes the code of the evaluation function very difficult to read and we can not say for sure if everything is correct  
f. e. if a move ends the game we will not notice it

.LP
Simulating the moves in the evaluation function is not what you usually do because the search simulates all moves already.  
But we only go down one branch of the tree and do it much faster because we only track our side.

Apart from that
We save the
.CW getMoves
results in a hashmap using the previous moves (saved in one long) as key.  
This allows us to put the
.CW bestMoves
in the front for more pruning.

In addition we change the depth depending on how promising the move is by adding i/2 to the current depth for a move ordered at position i.

All in all we used the given
.CW MinMaxAgent
as a template and were heavily inspired by the winner of the last year.
(It says ``don't plagiarize`` in the template but the template given is just how minmax works)
.SH
Implementation
.LP
Java very similar to the given minmax template.
We implemented our own game logic because we saw the winners of last year did.
It is slightly faster than before but we are not 100% sure if everything works correctly.
We also store moves as bytes to save space in the hashmap.

.NH 2
na05kymu;ce84pegu (224867)
.SH
Idea
.LP
An agent that implements a MinMax algorithm with Alpha-Beta pruning.
We do move ordering with a shallow lookahead of 1 to help the pruning process.
As evaluation function/heuristic we used the store difference between us and the opponent.
We also check if at any point we have more than half the points and stop the search returning the move.
Also if the opponent has more than half the points we do the same thing.
.SH
Implementation
.LP
Agent is written in python using the template provided by
.CW https://wwwcip.cs.fau.de/~oc45ujef/ai/kalah/python.tar.gz .
It has kgp.py module implemented and ready for use to communicate with the server.
Also agent.py which holds the search algorithm as well as the evaluation function.
The agent is implemented as a generator that yields different search results with different depths.
.NH 2
jo43beqy (224931)
.SH
Idea
.LP
I tried to make a move tree which always expands (thinking on opponent time) and updates when the oponent makes a move to make irrelevant brachen inacitve.
Which Node is expanded depends on prio which is combination of eval score and depth
The idea was to make eval a dot product of the state with weights and use deep learning techniques to get the best values for that.

However i could not get that all to work in time so I actually did a basic minmax.
.SH
Implementation
.LP
Python similar to the provided template.

.NH 2
EdizDerBreite (224952)
.SH
Idea
.LP
Minmax with alpha-beta-pruning.
.SH
Implementation
.LP
The agent was written in
.CW C++ ,
using the Python implemenation of the protocol and communicating over
.CW stdin
/
.CW stdout.