Instructions to use Zipeng365/WISP with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Scikit-learn
How to use Zipeng365/WISP with Scikit-learn:
from huggingface_hub import hf_hub_download import joblib model = joblib.load( hf_hub_download("Zipeng365/WISP", "sklearn_model.joblib") ) # only load pickle files from sources you trust # read more about it here https://skops.readthedocs.io/en/stable/persistence.html - Notebooks
- Google Colab
- Kaggle
Download scripts/build_search_best.py from Zipeng365/WISP: direct link, hf CLI and curl.
- Browser
- Download file 5.38 kB
-
https://huggingface.co/Zipeng365/WISP/resolve/main/scripts/build_search_best.py
- Command line
-
hf download hf://Zipeng365/WISP/scripts/build_search_best.py
-
curl -L -o build_search_best.py https://huggingface.co/Zipeng365/WISP/resolve/main/scripts/build_search_best.py
5.38 kB
| """Generate src/wisp/_data/search_best.json from results/search_selections.csv. | |
| Rule (validation only; no test metric is read): for each data stream take the 30 final | |
| recipes of the budget-329 searches (5 folds x 3 seeds x WISP-Random/WISP-Evolution). | |
| Each final recipe is the candidate with the highest validation utility in its search. | |
| The stream's best recipe is the recipe selected most often; ties are broken by the | |
| number of folds it covers, then by the canonical JSON text (deterministic, never needed | |
| for the released files). Only ``canonical_recipe``, ``dataset_id``, ``fold``, | |
| ``model_seed``, ``method_id`` and ``budget`` are read. | |
| Usage: python scripts/build_search_best.py [--check] | |
| """ | |
| from __future__ import annotations | |
| import argparse | |
| import csv | |
| import json | |
| from collections import defaultdict | |
| from pathlib import Path | |
| ROOT = Path(__file__).resolve().parents[1] | |
| SOURCE = ROOT / "results" / "search_selections.csv" | |
| TARGET = ROOT / "src" / "wisp" / "_data" / "search_best.json" | |
| ALIASES = ROOT / "src" / "wisp" / "_data" / "dataset_aliases.json" | |
| FINAL_BUDGET = 329 | |
| # Canonical forms of catalog recipes, for the "same as" column. | |
| FIXED_CANONICAL = { | |
| "ss": {"name": "har_sculpt_forest_hmm", "params": {"classifier": "extratrees", "segment_sizes": [4, 8, 16], "symbolic_bins": 6, "use_hmm": True}}, | |
| "rc": {"name": "rocket_sketch_ridge_hmm", "params": {"classifier": "ridge", "n_kernels": 384, "use_hmm": True}}, | |
| "so": {"name": "symbolic_interval_forest", "params": {"use_hmm": True}}, | |
| "cse": {"name": "hybrid_cpu_ensemble", "params": {"use_hmm": True}}, | |
| "gis": {"name": "invariant_interval_shape_forest", "params": {"classifier": "extratrees", "n_shapelets": 32, "segment_sizes": [4, 8, 16], "symbolic_bins": 6, "use_hmm": True}}, | |
| "cis": {"name": "state_rocket_interval_hmm", "params": {"classifier": "ridge", "n_intervals": 24, "n_kernels": 384, "use_hmm": True}}, | |
| "ese": {"name": "posterior_cpu_ensemble", "params": {"use_hmm": True, "weights": [0.34, 0.33, 0.33]}}, | |
| } | |
| DISPLAY = {"adl": "ADL", "capture24": "CAPTURE-24", "domino": "DOMINO", "gotov": "GOTOV", "handy": "HANDY", | |
| "harmes_left": "HARMES left wrist", "harmes_right": "HARMES right wrist", "iuwds": "IUWDS", | |
| "paal_adl": "PAAL-ADL", "wisdm": "WISDM", "mhealth": "mHealth", "pamap2": "PAMAP2"} | |
| def _same_as(recipe): | |
| for rid, canon in FIXED_CANONICAL.items(): | |
| if recipe["name"] == canon["name"] and recipe["params"] == canon["params"]: | |
| return rid | |
| flipped = dict(canon["params"], use_hmm=False) | |
| if recipe["name"] == canon["name"] and recipe["params"] == flipped: | |
| return f"{rid} with decoder='none'" | |
| return None | |
| def build(): | |
| aliases = json.loads(ALIASES.read_text()) | |
| short = {} | |
| for name, full in aliases.items(): | |
| short.setdefault(full, name) # first alias listed is the primary short name | |
| runs = defaultdict(list) | |
| with SOURCE.open(newline="") as f: | |
| for row in csv.DictReader(f): | |
| if int(float(row["budget"])) != FINAL_BUDGET: | |
| continue | |
| runs[row["dataset_id"]].append(row) | |
| out = [] | |
| for dataset_id in sorted(runs, key=lambda d: list(short).index(d)): | |
| rows = runs[dataset_id] | |
| if len(rows) != 30: | |
| raise ValueError(f"{dataset_id}: expected 30 final searches, found {len(rows)}") | |
| groups = defaultdict(list) | |
| for row in rows: | |
| groups[row["canonical_recipe"]].append(row) | |
| ranked = sorted(groups.items(), | |
| key=lambda kv: (-len(kv[1]), -len({r["fold"] for r in kv[1]}), kv[0])) | |
| text, chosen = ranked[0] | |
| runner_up = ranked[1][1] if len(ranked) > 1 else [] | |
| recipe = json.loads(text) | |
| name = short[dataset_id] | |
| out.append({ | |
| "id": f"best:{name}", | |
| "family": recipe["name"], | |
| "params": recipe["params"], | |
| "paper_name": f"WISP search best recipe ({DISPLAY[name]})", | |
| "source": "search_best", | |
| "provenance": { | |
| "dataset_id": dataset_id, | |
| "selected": len(chosen), | |
| "searches": len(rows), | |
| "folds_covered": len({r["fold"] for r in chosen}), | |
| "folds_total": len({r["fold"] for r in rows}), | |
| "cells": sorted([f"{r['method_id']}/fold{r['fold']}/seed{r['model_seed']}" for r in chosen]), | |
| "runner_up_selected": len(runner_up), | |
| "same_as": _same_as(recipe), | |
| "selection_rule": ("most frequently selected final recipe among the 30 budget-329 searches " | |
| "(validation utility only); ties by folds covered"), | |
| "source_file": "results/search_selections.csv", | |
| }, | |
| }) | |
| return out | |
| def main(): | |
| ap = argparse.ArgumentParser() | |
| ap.add_argument("--check", action="store_true", help="fail if the committed file differs") | |
| a = ap.parse_args() | |
| text = json.dumps(build(), indent=1, sort_keys=False) + "\n" | |
| if a.check: | |
| if TARGET.read_text() != text: | |
| raise SystemExit("search_best.json is out of date; rerun scripts/build_search_best.py") | |
| print("search_best.json is up to date") | |
| return | |
| TARGET.write_text(text) | |
| print(f"wrote {TARGET.relative_to(ROOT)}") | |
| if __name__ == "__main__": | |
| main() | |