Hello.
I am a Japanese Nintendo Cup 2000(NC2000) specialist.
And I know the basics about the international Gen2OU.
Strategy of Pokemon can be developed by programming.
You know that the Damage Calculator is developing the strategy.
(This is being developed in the same way both in Japan and internationally)
We will deal with more advanced topics here.
I write about the current situation and prospects of Gen2 in Japan.
If you know international situation and post it, I would be grateful.
In particular, I see the potential for many replays stored in Showdown.
(In Japanese Gen2 Simulator(Gym Leader's Castle), there is a culture that replays cannot be published without the consent of players, so there are only a few replays)
[Japanese Gen2 situation and prospects]
1. Goodness of Japanese Gen2 for programmers
In Japan, Gen2 Pokemon offers programmers the following ideal situations:
2. Current situation of Gen2 in Japan
Gym Leader's Castle has a long history, but it is not open source, it has some unique specifications, and the server that runs it is also poor.
This is a constraint for programmers, so I created and published Java source code(PBSTools) that reproduced the business logic of Gym Leader's Castle.
Currently, research by PBSTools is underway.
PBSTools allows programmers to write their own thinking routines and has the ability to automatically game 10,000 times according to those routines.
Research using this function has started in 2019 in Japan.
This is a primitive attempt, not something that can be called AI, but nevertheless provided the following insights:
3. Prospect of Gen2 in Japan
The attempts mentioned above have been successful, but there are also problems.
The essence of the problem is that creating a thinking routine is manual.
This raises the following specific issues:
There are several programmers in Japan who try to do that.
Attempts with AI have not yet to generate useful insights, but this attempt has potential for the future.
I am a Japanese Nintendo Cup 2000(NC2000) specialist.
And I know the basics about the international Gen2OU.
Strategy of Pokemon can be developed by programming.
You know that the Damage Calculator is developing the strategy.
(This is being developed in the same way both in Japan and internationally)
We will deal with more advanced topics here.
I write about the current situation and prospects of Gen2 in Japan.
If you know international situation and post it, I would be grateful.
In particular, I see the potential for many replays stored in Showdown.
(In Japanese Gen2 Simulator(Gym Leader's Castle), there is a culture that replays cannot be published without the consent of players, so there are only a few replays)
[Japanese Gen2 situation and prospects]
1. Goodness of Japanese Gen2 for programmers
In Japan, Gen2 Pokemon offers programmers the following ideal situations:
- In Japan, the internet was widespread at the time of Gen2, so the simulator(Gym Leader's Castle) game was held in 2001. It is still in use and has accumulated knowledge over the long years.
(At the time of Gen1, the internet was not widespread yet, so no simulator was made in Japan) - The regulations based on NC2000 is consistently the mainstream in Japan, and there are few revisions to the regulations. And many specifications are known. Programmers do not have to worry about uncertain regulations or specifications.
- There are 30 to 50 active NC2000 specialists in Japan. And more than 10 of them are veterans with more than 10 years of experience. The results of programming can be checked by them.
(Please note that specialists cannot be trained immediately) - For programmers, the theme of Pokemon gives them insight into situations about each players deciding at the same time, and luck-related situations. From that perspective, Gen2, which has fewer elements than other generations, is a good place to start. The knowledge gained from Gen2 can also be applied to later generations (and even things other than Pokemon).
2. Current situation of Gen2 in Japan
Gym Leader's Castle has a long history, but it is not open source, it has some unique specifications, and the server that runs it is also poor.
This is a constraint for programmers, so I created and published Java source code(PBSTools) that reproduced the business logic of Gym Leader's Castle.
Currently, research by PBSTools is underway.
PBSTools allows programmers to write their own thinking routines and has the ability to automatically game 10,000 times according to those routines.
Research using this function has started in 2019 in Japan.
This is a primitive attempt, not something that can be called AI, but nevertheless provided the following insights:
- In the face-on-face of L55 Return/Fire Punch/Curse/Milk Drink Polkadot Bow Miltank vs L55 Double-Edge/Curse/SleepTalk/Rest Leftovers Snorlax, it was revealed that the winning percentage of Miltank is about 54%. It was also revealed that if Snorlax is not Sleeping, Miltank should prioritize attacking rather than fully recovering HP, and if Snorlax is Sleeping, Miltank should prioritize fully recovering HP.
(Source : https://seesaawiki.jp/pbs-thread/d/...%f3%a5%af%a4%ce%c2%d0%bf%b2%b8%c0%a5%ab%a5%d3) - In the face-on-face of L55 Double-Edge/Earth Quake/Curse/Rest Leftovers Snorlax vs L55 Thunder Bolt/Hidden Power Ice/Swagger/Substitute Leftovers Zapdos, it was revealed that the winning percentage of Zapdos is about 27%. It was also revealed that Zapdos should not use Substitute.
(Source : https://seesaawiki.jp/pbs-thread/d/...%bc%a4%ce%a5%bf%a5%a4%a5%de%a5%f3%be%a1%ce%a8) - In some face-on-face cases other than the above, it clarified the winning percentage and the action that should be done. And while this is not really underway, it also know that a similar approach is possible in the complex cases of involving Switch.
3. Prospect of Gen2 in Japan
The attempts mentioned above have been successful, but there are also problems.
The essence of the problem is that creating a thinking routine is manual.
This raises the following specific issues:
- Each attempt takes time. Therefore, the amount of results is small. And in the game, it can only calculate the winning percentage in a very simple case where players have used the same move continuously at face-on-face.
- This attempt requires experience of competitive NC2000 to create reasonable routines, and Java skills. Therefore, the number of people who can do it is very limited.
(To be honest, I am the only one who can do it) - Thinking routines are created by humans and cannot be routines that exceed human limits.
There are several programmers in Japan who try to do that.
Attempts with AI have not yet to generate useful insights, but this attempt has potential for the future.