or run

tessl search
Log in

Version

Workspace
tessl
Visibility
Public
Created
Last updated
Describes
pypipkg:pypi/scikit-learn@1.7.x

evals

scenario-1

rubric.jsontask.md
tile.json

tessl/pypi-scikit-learn

tessl install tessl/pypi-scikit-learn@1.7.0

A comprehensive machine learning library providing supervised and unsupervised learning algorithms with consistent APIs and extensive tools for data preprocessing, model evaluation, and deployment.

Agent Success

Agent success rate when using this tile

87%

Improvement

Agent success rate improvement when using this tile compared to baseline

0.99x

Baseline

Agent success rate without this tile

88%

task.mdevals/scenario-1/

Probabilistic Reliability Suite

Tools for uncertainty-aware regression, density-based anomaly scoring, and calibrated classification on tabular data.

Capabilities

Uncertainty-aware regression

  • Fitting on smooth one-dimensional data returns mean predictions within 0.05 of expected values and standard deviations above zero for unseen points @test
  • Lowering the smoothness setting produces predictions that fluctuate more across nearby points than when using a higher smoothness value on the same dataset @test

Density-based anomaly scoring

  • A model trained on a single cluster assigns higher average log-likelihood to in-cluster samples than to a distant outlier cluster @test

Probability calibration

  • Calibrated probabilities on a held-out set reduce Brier score compared to raw base model scores while preserving predicted labels @test
  • Calibrated probabilities sum to one for each sample and stay within [0, 1] bounds @test

Implementation

@generates

API

from typing import Iterable, Tuple

ArrayLike = Iterable[Iterable[float]]
VectorLike = Iterable[float]


class ProbabilisticSuite:
    def fit_regressor(self, features: ArrayLike, targets: VectorLike, smoothness: float = 1.0) -> None:
        """Train a probabilistic regressor that returns mean and uncertainty estimates."""

    def predict_regressor(self, features: ArrayLike) -> Tuple[VectorLike, VectorLike]:
        """Return mean predictions and standard deviations for each sample."""

    def fit_density_estimator(self, features: ArrayLike, components: int = 2) -> None:
        """Train a mixture-based density model capable of per-sample log-likelihood scoring."""

    def score_density(self, features: ArrayLike) -> VectorLike:
        """Return log-likelihood scores for each sample."""

    def fit_calibrated_classifier(self, features: ArrayLike, labels: VectorLike, cv: int = 3, method: str = "sigmoid") -> None:
        """Train a base classifier and wrap it with probability calibration using cross-validation."""

    def predict_calibrated_proba(self, features: ArrayLike) -> ArrayLike:
        """Return calibrated class probabilities for each sample."""

    def predict_calibrated_labels(self, features: ArrayLike) -> VectorLike:
        """Return class labels based on the calibrated probabilities."""

Dependencies { .dependencies }

scikit-learn { .dependency }

Provides probabilistic regression, mixture-based density estimation, and probability calibration utilities.