Skip to content

aspective

Peer-reviewed variants of classic estimators that never made it into scikit-learn, with the scikit-learn API.

Naive Bayes, LDA, k-NN, CART, random forests, AdaBoost, the lasso, logistic regression and Isolation Forest are the workhorses of applied machine learning. Over decades, the statistics and machine learning literature published variants of each that fix a known weakness, and compared them carefully with the original. Most of them never reached a mainstream Python library. aspective collects them.

  • From the literature

    Every estimator comes from a paper in a strong venue that compared it with its vanilla version in a principled way, and its docstring cites that paper and says when to prefer it.

  • Benchmark-gated

    An estimator is only included if, on dozens of public datasets, it is statistically non-inferior to the model it improves on. Candidates that fail are removed, and the benchmarks say which.

  • Drop-in

    Supervised estimators pass scikit-learn's check_estimator, so they work with pipelines, grid search and cross-validation. Anomaly detectors are genuine PyOD detectors when PyOD is installed.

  • Easy to choose

    Start from the model you already use, and the guide lists its variants and when each helps.

A first example

from sklearn.datasets import load_wine
from sklearn.model_selection import GridSearchCV
from aspective.discriminant_analysis import RegularizedDiscriminantAnalysis

X, y = load_wine(return_X_y=True)
search = GridSearchCV(
    RegularizedDiscriminantAnalysis(),
    {"lam": [0, 0.25, 0.5, 0.75, 1], "gamma": [0, 0.1, 0.5]},
).fit(X, y)
print(search.best_params_, search.best_score_)

lam=0, gamma=0 is QDA and lam=1, gamma=0 is LDA, so the grid search picks between them and everything in between.

Next steps