skycmd-games / README.md
maxgfr's picture
skycmd 6cbcce8c: skycmd game decisions
fa4af24 verified
|
Raw History Blame Contribute Delete
2.09 kB
---
license: apache-2.0
language:
- en
pretty_name: skycmd game decisions
task_categories:
- text-classification
tags:
- skycmd
- typed-decisions
- system-one
configs:
- config_name: dodge
data_files: data/dodge.parquet
- config_name: doom
data_files: data/doom.parquet
- config_name: flappy
data_files: data/flappy.parquet
- config_name: rescue
data_files: data/rescue.parquet
---
# skycmd game decisions
Decision questions from the four skycmd control games (Flappy Drone, Rescue Run, Obstacle Dodge, a Doom-like shooter), labelled by the games' exact oracles. Generated by our Rust game cores: no third-party data.
Part of [skycmd](https://github.com/maxgfr/skycmd), tiny decision models trained from scratch in Rust that run in the
browser. Written by `sky-games-gen --game all --n 60000 --seed 1, and --game doom --n 100000 --seed 1 (scripts/prepare-data.sh)` at commit [`6cbcce8c`](https://github.com/maxgfr/skycmd/commit/6cbcce8cbbbf27e8e952e61b6cb1ed03ee1fca58).
Licences of every set: [DATA_LICENSES.md](https://github.com/maxgfr/skycmd/blob/main/DATA_LICENSES.md).
Each row is one typed decision question (docs/SPEC.md):
| field | type | meaning |
|---|---|---|
| `src` | string | source |
| `t` | string | `choice`, `score` or `noul` |
| `ins` | string | the question |
| `opts` | list of strings | the options, rendered as text |
| `state` | string | the state the question is about |
| `target` | list of floats | a probability per option (soft or one-hot) |
| `split` | string | `train`, `val` or `test` |
## Licence
Apache-2.0: generated by skycmd's own code, no third-party data.
## Files
| file | rows |
|---|---:|
| `data/dodge.parquet` | 60,000 |
| `data/doom.parquet` | 100,000 |
| `data/flappy.parquet` | 60,000 |
| `data/rescue.parquet` | 60,000 |
```python
from datasets import load_dataset
ds = load_dataset("maxgfr/skycmd-games", "dodge")
```
skycmd wrote no row with a language model: the targets come from the upstream labels (some upstream
sets carry their own teacher soft labels, see their cards), the games' oracles or our generators.