MarkovJuniorWeb

Typescript version of MarkovJunior, runs in browser (also in node.js).

• Everything have been implemented including isometric rendering, exporting the output as a .vox file, and node tree visualization.
• Every model from the original repository can be loaded with this project, but the output would be different due to different random seed implementation (dotnet builtin vs seededrandom).

Development

Install dependensies:

npm i

Start development server on localhost:

npm start

Build static site:

npm build

Run in node (writes result to /output):

npm run cli

TODO

• node editor
• import from & export to xml
• voxel editor
• file system on the web
• optimization

Performance

This port is around 2x slower than the original repo (JS vs C#), but it doesn't affect the page much; even with 200 steps per frame there's hardly any FPS drop on most models. However, the slowdown is quite noticable on computation expensive calculations, e.g. uni/bi-direction inference.

Random Notes

SokobanLevel1 takes ~10 seconds for the original C# code on my pc to reach the desired state, while it takes 20+ seconds on the web. I've tried JIT/unroll the rules into webassembly with generated AssemblyScript and it actually works: it gains a x2 speedup and the performance almost match the native C# version. The only problem is the load & compile time is terrilbe and it's incredibly hard to debug WebAssembly. I rolled back the commits on main and put the experimental stuff in the optimization branch, but I'm still pretty proud of this MarkovJunior rules -> AssemblyScript -> Wasm JIT compiler I wrote.

Update: I wrote a precompiled wasm version and it works fine, and the runtime is reduced from 20+ seconds on SokobanLevel1 to ~13 seconds (not too bad I guess ¯\_(ツ)_/¯ ).