wyj/katrain-qt
1
0 Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

rank model

Browse Source
This commit is contained in:
Sander Land committed 2020-07-05 22:24:17 +02:00
1 parent 32074d6616
commit f4f8f61084
6 files changed
+128 -36
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

README.md
+8 -3
View File
@@ -76,13 +76,18 @@ on the board, or not output details for them in SGFs,you can do so under 'Config
This section describes the available AIs, with strength based on their current OGS rankings using the default settings.
In the 'AI settings', settings with calibration are at the top and have a lighter color, changing these will show an estimate of rank.
This estimate is reasonably accurate as long as you have not changed the other settings.
* Recommended options for serious play include:
* **[9p+]** **KataGo** is full KataGo, above professional level. The analysis and feedback given is always based on this full strength KataGo AI.
* **[15k - 3d]** **Calibrated Rank Bot** was calibrated on various bots (e.g. GnuGo and Pachi at different strength settings) to play a balanced game from the opening to the endgame without making serious (DDK) blunders. Further discussion can be found on [this](https://github.com/sanderland/katrain/issues/44) thread.
* **[15k - 3d]** **Calibrated Rank Bot** was calibrated on various bots (e.g. GnuGo and Pachi at different strength settings) to play a balanced
game from the opening to the endgame without making serious (DDK) blunders. Further discussion can be found
[here](https://github.com/sanderland/katrain/issues/44) and [here](https://github.com/sanderland/katrain/issues/74).
* **[~5k]** **ScoreLoss** is KataGo analyzing as usual, but
choosing from potential moves depending on the expected score loss, leading to a varied style with mostly small mistakes.
* **[~4d]** **Policy** uses the top move from the policy network (it's 'shape sense' without reading).
* **[~4k]** **Policy Weighted** picks a random move weighted by the policy, leading to a varied style with mostly small mistakes, and occasional blunders due to a lack of reading.
* **[~5d]** **Policy** uses the top move from the policy network (it's 'shape sense' without reading).
* **[12k - 2d]** **Policy Weighted** picks a random move weighted by the policy, leading to a varied style with mostly small mistakes, and occasional blunders due to a lack of reading.
* **[~8k]** **Blinded Policy** picks a number of moves at random and play the best move among them, being effectively 'blind' to part of the board each turn.
* Options that are more on the 'fun and experimental' side include:
* Variants of **Blinded Policy**, which use the same basic strategy, but with a twist.: