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

Post edit history

Want to make a deep learning AI for Zero-K

To display differences between versions, select one or more edits in the list using checkboxes and click "diff selected"
Post edit history
Date Editor Before After
10/29/2026 5:30:43 AMGBrankPresstabstart before revert after revert
Before After
1 Consider using my replay scraper+parser ( if it still works) : 1 Consider using this replay scraper+parser ( if it still works) :
2 https://github.com/pts1667/zkscraper 2 https://github.com/pts1667/zkscraper
3 \n 3 \n
4 I have a "remote controller" AI as well. You will need to compile for the latest spring version: 4 There is a "remote controller" AI as well. You will need to compile for the latest spring version:
5 https://github.com/pts1667/ControllerAI 5 https://github.com/pts1667/ControllerAI
6 \n 6 \n
7 Ofc, this means I've tried something similar in the past. 7 Ofc, this means I've tried something similar in the past.
8 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). 8 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).
9 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. 9 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.
10 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. 10 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.
11 Make sure to apply greater weighting to real replays/RL sims. 11 Make sure to apply greater weighting to real replays/RL sims.
12 \n 12 \n
13 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. 13 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.