|
2 | 2 |
|
3 | 3 | The scripts in this directory reproduce benchmark results used in the paper. |
4 | 4 |
|
5 | | -## Additional dependencies |
| 5 | +## Benchmark environment |
6 | 6 |
|
7 | 7 | The benchmark script (`benchmark.py`) additionally imports: |
8 | 8 |
|
9 | 9 | - `xgboost` |
10 | 10 | - `catboost` |
11 | 11 | - `tabpfn` |
12 | 12 |
|
13 | | -These are **benchmark-only** dependencies. They are intentionally not required by |
14 | | -the core `sklearn-contrib-bde` package and are therefore not listed under the |
15 | | -main project dependencies. |
| 13 | +These are **benchmark-only** dependencies. They are intentionally kept out of |
| 14 | +the core `sklearn-contrib-bde` package and provided through the optional pixi |
| 15 | +`benchmark` environment instead. |
16 | 16 |
|
17 | 17 | ## Setup |
18 | 18 |
|
19 | 19 | Run from the repository root unless noted otherwise. |
20 | 20 |
|
21 | | -1. Create/install the base project environment as specified in the main project |
22 | | -documentation. |
| 21 | +1. Install the benchmark environment: |
23 | 22 |
|
24 | | -2. Install benchmark-only Python packages into the pixi environment: |
| 23 | +```bash |
| 24 | +pixi install -e benchmark |
| 25 | +``` |
| 26 | + |
| 27 | +This installs the locked benchmark dependencies, including `pip`, `xgboost`, |
| 28 | +`catboost`, and `tabpfn`; no manual `pip install` or `ensurepip` step is needed. |
| 29 | + |
| 30 | +2. Run a benchmark through the pixi task. For example: |
25 | 31 |
|
26 | 32 | ```bash |
27 | | -pixi run python -m pip install xgboost catboost tabpfn |
| 33 | +pixi run -e benchmark benchmark --dataset airfoil --models linear rf --n-runs 5 |
28 | 34 | ``` |
29 | 35 |
|
30 | | -3. Run the benchmark script `paper/scripts/run_bench.sh` to reproduce the benchmark. |
| 36 | +The task invokes `paper/scripts/run_bench.sh`. The wrapper resolves its paths |
| 37 | +relative to its own location, so this equivalent command also works from the |
| 38 | +repository root: |
| 39 | + |
| 40 | +```bash |
| 41 | +pixi run -e benchmark bash paper/scripts/run_bench.sh \ |
| 42 | + --dataset airfoil --models linear rf --n-runs 5 |
| 43 | +``` |
0 commit comments