Loading...
  OR  Zero-K Name:    Password:   

Want to make a deep learning AI for Zero-K

6 posts, 378 views
Post comment
Filter:    Player:  
sort
I want to make a deep learning AI for Zero-K (with deepseek programming) (just trying vibe coding and deep learning development).

I want to know if this is allowed and whether my needs can be met.

Form



An player with account, using LuaUI and luasocket to connect with AI.

Ability



I may implement these step-by-step:

1: AI player host a lobby and accept any player, player !start to challenge AI. And AI play matchmaking.

2: AI can join TAW or Palladium (if it can have 2200+ elo) and play multiplayer games.

3: AI can chat.

Need



Download a lot of replays, Pro 1v1 Host and 1v1 MM.
+2 / -0

24 days ago
I see no reason not to let it have a lobby where people can choose to play against it and it receives an elo score.

I however don't want to play against AI in my taw or palla experience, and certainly don't want an AI chatbot in my games.
+5 / -0

24 days ago
An ai that gives tips post games would be nice.
+0 / -0
Consider using this replay scraper+parser (if it still works):
https://github.com/pts1667/zkscraper

There is a "remote controller" AI as well. You will need to compile for the latest spring version:
https://github.com/pts1667/ControllerAI

Ofc, this means I've tried something similar in the past.
The biggest issue is you can't effectively use RL or parse replays quickly due to Spring's massive CPU+memory requirements (relative to most other RL environments).
If I were to try this again, I would start with creating a simplified environment (engine, game code, etc.) so that I can run a large amount of RL instances in parallel, mixing in say ~1% actual matches (vs CPU/other AI). I would begin training on raw replays first then proceed with RL in a second training stage.
You will need a LOT of replays. A small amount like 1000, 2000 won't do. You could mix in replays from the simplified environment.
Make sure to apply greater weighting to real replays/RL sims.

Consider fine-tuning an existing LLM model. Not something I've tried, but I'd look into replacing the LM head with dedicated rts output controls, if that makes sense.
+1 / -0
19 days ago
So can I download replays now?
+0 / -0
Yes, it lets you download and parse replays. I've used it a while ago, it's a scraper so it can be pretty brittle. If the replays page changed recently (zero-k did move to a new host) it could break... but nothing you can't ask AI to apply a small fix for.

Make sure you are considerate with rate limits, the default rate limits are kind of slow but it might be OK to increase now that they've moved to a new host. I haven't tested the scraper on the new host.
+0 / -0