search

datamllab / rlcard

Reinforcement Learning / AI Bots in Card (Poker) Games - Blackjack, Leduc, Texas, DouDizhu, Mahjong, UNO.
http://www.rlcard.org
MIT License
2.86k stars 618 forks source link

Difference in doudizhu between source code and documents #157

Closed fecet closed 4 years ago

fecet commented 4 years ago

The documents say The size of the action space of Dou Dizhu is 33676. But the json file in games/doudizhu/jsondata/card_type.json only contains 27471, more detail:

trio_solo_chain_5 10388 trio_solo_chain_4 7184 trio_solo_chain_3 3282 four_two_solo 1326 trio_pair_chain_3 1200 trio_pair_chain_4 1134 trio_solo_chain_2 968 four_two_pair 858 trio_pair_chain_2 605 trio_solo 182 trio_pair 156 solo 15 pair 13 trio 13 bomb 13 trio_chain_2 11 trio_chain_3 10 pair_chain_3 10 pair_chain_4 9 trio_chain_4 9 trio_chain_5 8 pair_chain_5 8 solo_chain_5 8 pair_chain_6 7 trio_chain_6 7 solo_chain_6 7 solo_chain_7 6 pair_chain_7 6 pair_chain_8 5 solo_chain_8 5 solo_chain_9 4 pair_chain_9 4 pair_chain_10 3 solo_chain_10 3 solo_chain_11 2 rocket 1 solo_chain_12 1

seems there is something wrong in plane case.

CaoYuanpu commented 4 years ago

@fecet Thanks for letting us know. I have updated the actions in documents. Now the size of the action space of Dou Dizhu is 27472 with a "pass" action which is not in json file.

fecet commented 4 years ago

@fecet Thanks for letting us know. I have updated the actions in documents. Now the size of the action space of Dou Dizhu is 27472 with a "pass" action which is not in json file.

But this is still less than 33676 mentioned in documents

Type Number of Actions Number of Actions after Abstraction Action ID
Solo 15 15 0-14
pair 13 13 15-27
Trio 13 13 28-40
Trio with single 182 13 41-53
Trio with pair 156 13 54-66
Chain of solo 36 36 67-102
Chain of pair 52 52 103-154
Chain of trio 45 45 155-199
Plane with solo 24721 38 200-237
Plane with pair 6552 30 238-267
Quad with solo 1339 13 268-280
Quad with pair 1014 13 281-293
Bomb 13 13 294-306
Rocket 1 1 307
Pass 1 1 308
Total 33676 309
CaoYuanpu commented 4 years ago
@fecet The documents(https://github.com/datamllab/rlcard/blob/master/docs/games.md#dou-dizhu) in github have been updated, and we will soon update the documents in http://rlcard.org/games.html#dou-dizhu Type Number of Actions Number of Actions after Abstraction Action ID
Solo 15 15 0-14
pair 13 13 15-27
Trio 13 13 28-40
Trio with single 182 13 41-53
Trio with pair 156 13 54-66
Chain of solo 36 36 67-102
Chain of pair 52 52 103-154
Chain of trio 45 45 155-199
Plane with solo 21822 38 200-237
Plane with pair 2939 30 238-267
Quad with solo 1326 13 268-280
Quad with pair 858 13 281-293
Bomb 13 13 294-306
Rocket 1 1 307
Pass 1 1 308
Total 27472 309
hsywhu commented 4 years ago

Related docs on the official website have been updated. http://rlcard.org/games.html#action-abstraction-of-dou-dizhu