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 validation/software_validation_0.4.0.json from Zipeng365/WISP: direct link, hf CLI and curl.
- Browser
- Download file 4.56 kB
-
https://huggingface.co/Zipeng365/WISP/resolve/main/validation/software_validation_0.4.0.json
- Command line
-
hf download hf://Zipeng365/WISP/validation/software_validation_0.4.0.json
-
curl -L -o software_validation_0.4.0.json https://huggingface.co/Zipeng365/WISP/resolve/main/validation/software_validation_0.4.0.json
4.56 kB
| { | |
| "version": "0.4.0", | |
| "status": "passed", | |
| "tested_commit_note": "Test runs used a wheel built from a development commit (git archive), installed into a fresh virtual environment, with the tests run against the installed wheel (import paths checked by tests/test_installed.py). That commit's src/, tests/, setup.py and pyproject.toml are identical to the release commit (git tree of src/ f9ebebaddec0d4068511c71c5d8701a1cce07c95, tests/ 4e8692d44d9aa462379f4e2b4a03385b2f8c5c74); the release commit adds documentation, benchmark tables, the failure log and these records. An earlier development state passed the same suites (Slurm 61882979, 61882980).", | |
| "environment": { | |
| "python": "3.12.13", | |
| "numpy": "2.3.5", | |
| "scipy": "1.17.1", | |
| "scikit-learn": "1.7.2", | |
| "joblib": "1.5.3", | |
| "threadpoolctl": "3.6.0", | |
| "pytest": "9.0.3", | |
| "blas_threads": 1 | |
| }, | |
| "runs": [ | |
| { | |
| "name": "full suite, C kernels compiled", | |
| "extras": "optuna, torch (CPU), skorch, pandas", | |
| "equivalence_tests": true, | |
| "hub_test": "not requested", | |
| "slurm_job_id": "61885548", | |
| "passed": 288, | |
| "failed": 0, | |
| "skipped": 1, | |
| "subtests_passed": 239 | |
| }, | |
| { | |
| "name": "full suite, no C compiler (pure NumPy paths)", | |
| "extras": "optuna, torch (CPU), skorch, pandas", | |
| "equivalence_tests": true, | |
| "hub_test": "not requested", | |
| "slurm_job_id": "61886087", | |
| "skip_reasons": "14 tests of the optional native kernels; the Hub download test", | |
| "passed": 274, | |
| "failed": 0, | |
| "skipped": 15, | |
| "subtests_passed": 95 | |
| } | |
| ], | |
| "equivalence": "Equivalence tests E1-E7 (tests/test_equivalence.py, decoder='recipe') on ADL-Wrist fold 0 give a report identical to the 0.3.0 report apart from timings (59 records), with and without the C kernels.", | |
| "funnel_record_reproduction": "WISPSearch(strategy='funnel', decoder='recipe', n_jobs=2) with each export's sampling rate and study setting on all 150 cells of the funnel benchmark record (ten primary streams x 5 folds x seeds 42-44; Intel Xeon Platinum 8570, 2 cores, one BLAS thread): 300 of 300 controller results have the same winner, identical test class scores and decoded labels and identical metrics; the recomputed nine-view averages (Random 69.22, Evolution 69.30 macro-F1) and per-dataset tables are identical to the record (Slurm 61885546; benchmarks/results/funnel_record_check.md).", | |
| "funnel_without_c_kernels": "A build without a C compiler gave the same winners, test class scores and labels as the record on ADL-Wrist fold 1 seed 42 and fold 3 seed 44 (Slurm 61882552).", | |
| "funnel_external_harness": "The funnel through the package, called from an aeon/tsml-eval wrapper, against an independent earlier integration of the same method code on 8 datasets of the Multiverse archive (resample 0): BasicMotions, Epilepsy, JapaneseVowels, KERAAL-RTK (77 channels), IRDS-EFL (75), FaceDetection (144), and two cases where no candidate passed validation and on_no_candidate='default' fitted the default CIS recipe (KIMORE-LA-C, KINECAL-QSEO with 75 channels): test probabilities identical in all 16 of 16 controller results, the same winners (Slurm 61882968-61882975). Run with an earlier development build; the later code change adds only the optional recording descriptors, which this harness does not pass.", | |
| "saved_cells_sample": "benchmarks/check_against_saved_results.py (decoder='recipe') on ADL-Wrist fold 2, HANDY fold 1, PAAL-ADL fold 4 and DOMINO fold 3, seeds 42-44: the 7 fixed recipes and WISP-CIS without HMM with speed='exact', and the Ridge recipes with speed='fast' (156 records): all 156 observed metrics identical to the 0.3.0 all-cell check; all 96 exact-tier records equal the saved paper metrics (Slurm 61883165).", | |
| "clean_install": "The release wheel (dist/wisp_har-0.4.0-cp312-cp312-linux_x86_64.whl, sha256 dd6de3220be5dcf57c878a0e76a5e0d18e2b0cd065b0dac48953e5d454447e6b, built from the release commit) installed into a fresh virtual environment with environments/cpu.lock.txt, imported from site-packages with the four funnel kernels present; a synthetic WISPSearch('funnel') fit with the default decoder ran, and on ADL-Wrist fold 1 seed 42 both controllers matched the benchmark record (same winners, identical test class scores and labels) (Slurm 61936566).", | |
| "name_check": "Case-insensitive search of the repository tree and of the unpacked wheel for the internal version name of the source and its module names: no occurrence in any text file, file name or commit message." | |
| } | |