Getting started¶
The documentation site is built with MkDocs Material and reads the local Python modules directly from the repository root.
Install docs dependencies¶
Preview the site¶
Build a static version¶
What you will find¶
- A conceptual guide for the main noise models handled by the library.
- A dedicated page for TabPFN-based filtering and local explainability.
- An evaluation guide for 5-fold experiments, persisted noise masks, and continuous noise scores.
- An API reference generated directly from the public docstrings.
Minimal examples¶
Generate label noise and keep its true mask:
from noisers.funcs import urlf
y_noisy, noise_mask = urlf(y, noise_level=0.2, random_state=42, return_mask=True)
Run a filter in detection mode:
from filters import ENNFilter
filt = ENNFilter(action="detect")
filt.fit_resample(X, y_noisy)
report = filt.get_detection_report()
Combine multiple filters with FEF:
from filters import ENNFilter, CVCFFilter, FilterEnsembleFilter
fef = FilterEnsembleFilter(
base_filters=[("ENN", ENNFilter()), ("CVCF", CVCFFilter())],
strategy="majority",
)
X_clean, y_clean = fef.fit_resample(X, y_noisy)
Tip
Run MkDocs from the repository root so the local filters and cleaners packages can be imported without any extra packaging step.