The T-spin AI bot (MisaMino ver 1.4.5 1111)

[StS]

KoS mode makes me very happy. All-spins would be even more awesome, but I guess the AI is not made for all-spins?

[farter]

another new version is up, it supports internal lv8 vs lv8 now.

[erstwhile]

Any plans on make T-Spin AI do triple?

[farter]

Any plans on make T-Spin AI do triple?

it's capable to do that, however as the result of many test runs, it makes t3 very rarely. seems it's because of leading to bad shape which makes it hard to dig down.

[Draken]

Dang nice bot lol, only won once out of like 10.

[vipjun]

Challenge for misabot :

Since most of us cannot defeat AI 7+ at a 1:1 ratio
My challenge is to be able to defeat human player but playing slower.

So for starters Human plays 7 pieces, Misabot plays only 6 pieces.

[farter]

Challenge for misabot :

Since most of us cannot defeat AI 7+ at a 1:1 ratio
My challenge is to be able to defeat human player but playing slower.

So for starters Human plays 7 pieces, Misabot plays only 6 pieces.

there are already some ways of handicapping implemented, like no T-spin, no hold, set as "style" in misamino.ini.
and they are still hard to defeat with lv7-8 set.

[misakamm]

V1.4.5 AI infomation

T-Spin AI

Code Select Expand

   searching depth  searching nodes
LV0       0               ALL
LV1       1              4000
LV2       2              4000
LV3       3              4000
LV4       4              4000
LV5       5              5000
LV6       6              6000
LV7       6              8000
LV8       6             16000

searching depth means how many preview pieces it would see

in this version, you can set the pieces ratio in turnbase mode, @vipjun

[Integration]

It's 4000/5000/6000/8000/16000 nodes on the deepest level (search depth), isn't it?

I would like to know how you determine which nodes you examine. It's not completely random, right? I think it's clear how your algorithm works with 1 preview. Could you explain it on the basis of 2 previews (or give me a link to the chess system if it is too complicated)?

At first you create all nodes on the first level (i.e. you try to drop the first piece on all possible locations)? After that you create all 2nd level nodes which are children of the best rated 1st level node? Then you create all children of the 2nd best rated 1st level node and continue? After 4000 nodes you stop? At the end you always choose the 2nd level node with the best rating?

[farter]

It's 4000/5000/6000/8000/16000 nodes on the deepest level (search depth), isn't it?

I would like to know how you determine which nodes you examine. It's not completely random, right? I think it's clear how your algorithm works with 1 preview. Could you explain it on the basis of 2 previews (or give me a link to the chess system if it is too complicated)?

At first you create all nodes on the first level (i.e. you try to drop the first piece on all possible locations)? After that you create all 2nd level nodes which are children of the best rated 1st level node? Then you create all children of the 2nd best rated 1st level node and continue? After 4000 nodes you stop? At the end you always choose the 2nd level node with the best rating?

it works basically same as you said, but the limit is on the number of best nodes to iterate on each level, and that number seems to be an approximate count of all node created. (as what i heard from misakamm)

[Integration]

it works basically same as you said, but the limit is on the number of best nodes to iterate on each level, and that number seems to be an approximate count of all node created. (as what i heard from misakamm)

Understood. So the game completes one level (all possible spots or until level's node limit is reached) before it goes to the next level. First it tries to drop the current piece on all possible spots. Then it chooses the resulting playfield with the best rating and tries to drop the first preview piece on all possible spots. Then it chooses the resulting playfield with the second best rating and tries to drop the first preview piece on all possible spots. This is continued until 4000 fields emerged from 2 dropped pieces are created. Same procedure until 4000 fields emerged from 3 dropped pieces are created. And so on...

[VladtheImpala]

Have you tried programming in a "human" sense of consistency, i.e. once in a while it makes a misdrop(1 extra input). I think it would be interesting to see the ai's play if it occasionally had a miscue, and how it would effect its efficiency.

[morningpee]

I saw your perfect clear video when farter posted it to Hard Drop's video collection.

Very nice job!  Will you please make the perfect clear style available for download?

[Antifate]

Misa if you're watching - stop making the rooms passworded D:

Or at least tell me the password  

[misakamm]

Misa if you're watching - stop making the rooms passworded D:

Or at least tell me the password  

sorry, that room just is a test, actually bot is online with another name