sklplus

中文文档

sklplus 0.1.1 is a thin umbrella around scikit-learn-style tabular estimators. Most symbols are re-exports (often the same object as upstream). A few adapters and custom preprocessors fill gaps we care about when building Pipelines by hand.

It is for people who want from sklplus.… import … and standard fit / transform / predict composition. It is not AutoML: there is no get_model, no setup(), and no string-ID model factory.

Fits: classification, regression, clustering, and anomaly detection on tabular data, with preprocess / sampling / selection / search / metrics helpers re-exported under sklearn-like paths.

Does not fit (v0.1): time-series or NLP specialist modules. If you need PyCaret-style experiment sessions, use something else.

Requirements

• Python >=3.10
• scikit-learn>=1.3
• Default install also pulls: numpy, pandas, scipy, joblib, xgboost, lightgbm, catboost, imbalanced-learn, pyod, kmodes, category-encoders, feature-engine

Install

PyPI / import / GitHub repo name: sklplus (https://github.com/IncubatorShokuhou/sklplus).

pip install sklplus

From a clone (editable, with test tools):

pip install -e ".[dev]"

[dev] adds pytest>=7 and ruff.

Minimal usage

from sklplus.linear_model import LogisticRegression, Ridge
from sklplus.ensemble import RandomForestClassifier, XGBClassifier
from sklplus.preprocessing import (
    StandardScaler,
    CleanColumnNames,
    RareCategoryGrouper,
    DateFeatureExtractor,
    TargetLabelEncoder,
)
from sklplus.sampling import SMOTE
from sklplus.anomaly import IForest
from sklplus.pipeline import Pipeline, ImbPipeline
from sklplus.compose import ColumnTransformer

Re-exports aim to be identity with upstream when possible, e.g. sklplus.ensemble.RandomForestClassifier is sklearn.ensemble.RandomForestClassifier.

Boosting import paths

XGBClassifier / LGBMClassifier / CatBoostClassifier (and the regressors) are exported from both sklplus.ensemble and sklplus.xgboost / lightgbm / catboost. They are the same class object:

from sklplus.ensemble import XGBClassifier as A
from sklplus.xgboost import XGBClassifier as B
assert A is B

Resampling

Use ImbPipeline when a step calls fit_resample (e.g. SMOTE). Plain Pipeline is the sklearn one.

Examples

Scripts under examples/:

python examples/01_classification_pipeline.py

Custom preprocessors in 0.1.1

These live in sklplus.preprocessing (alongside re-exported scalers/encoders):

P0

• CleanColumnNames
• DateFeatureExtractor
• RareCategoryGrouper
• TargetLabelEncoder

P1

• GroupFeatures — row-wise aggregates over named column groups
• RemoveMulticollinearity — drop highly correlated numeric features (optional prefer-by-y)
• IterativeImputerPlus — IterativeImputer for numeric + most-frequent for categoricals
• TextEmbedder — BoW / TF-IDF on text columns, concatenated back

Sampling (not preprocessing): RemoveOutliers in sklplus.sampling is an imblearn-style sampler (fit_resample) for train-only outlier row drops — use with ImbPipeline.

Anomaly detectors under sklplus.anomaly are thin wrappers around pyod (IForest, LOF, …) so they can sit in a sklearn Pipeline.

Known gaps

Still open: Optuna/skopt search wrappers; broader check_estimator coverage; a docs site. P1 custom preprocessors / RemoveOutliers sampler are implemented.

License: MIT (see LICENSE).

Contributing: TODO.