scikit-learn

scikit-learn

Machine learning in Python with scikit-learn covering classification, regression, clustering, dimensionality reduction, pipelines, and hyperparameter tuning.

What it does

This skill provides three CLI-ready scripts for common machine learning workflows. The demo script fits a log-log linear regression on synthetic or upstream data, returning R-squared, slope, intercept, and a scaling exponent (alpha). It accepts upstream data via --input-json for integration with other skills in agent pipelines.

The classification pipeline script demonstrates a complete supervised learning workflow: preprocessing with ColumnTransformer (numeric scaling, categorical one-hot encoding), model comparison via 5-fold cross-validation (Logistic Regression, Random Forest, Gradient Boosting), hyperparameter tuning with GridSearchCV, and full test-set evaluation with accuracy, precision, recall, F1, ROC AUC, confusion matrix, and feature importances.

The clustering analysis script compares K-Means, Agglomerative Clustering, Gaussian Mixture, and DBSCAN. It includes an elbow method and silhouette analysis for optimal K selection, evaluation with silhouette score, Calinski-Harabasz index, and Davies-Bouldin index, plus PCA-based 2D visualization of cluster assignments.

Setup

cd scikit-learn
python3 -m venv .venv && source .venv/bin/activate && pip install scikit-learn numpy pandas matplotlib -q

Environment variables

None.

Usage

Input

python3 scripts/demo.py --query "SSTR2 binding potency scaling" --format json

Output

{
  "topic": "SSTR2 binding potency scaling",
  "model": "LinearRegression",
  "data_source": "synthetic",
  "n_observations": 40,
  "x_label": "log10(concentration)",
  "y_label": "log10(response)",
  "coefficients": {"intercept": 3.456, "slope": -1.412},
  "r2": 0.998,
  "r_squared": 0.998,
  "alpha": 1.412,
  "scaling_exponent": 1.412
}

CLI flags

demo.py

Flag	Description	Default
--query, -q	Topic to analyse	"general trend"
--format, -f	Output format (summary or json)	summary
--input-json	JSON with upstream data {"rows": [{"x":..,"y":..}]}	""
--describe-schema	Print expected input-json schema and exit	off

classification_pipeline.py

Run directly with python3 scripts/classification_pipeline.py (uses Breast Cancer Wisconsin dataset as example).

clustering_analysis.py

Run directly with python3 scripts/clustering_analysis.py (uses Iris dataset and synthetic blobs as examples).

Scripts

Script	Purpose
demo.py	CLI entrypoint -- log-log linear regression with synthetic or upstream data
classification_pipeline.py	Full classification workflow: preprocessing, model comparison, tuning, evaluation
clustering_analysis.py	Clustering comparison: K-Means, Agglomerative, GMM, DBSCAN with evaluation

Dependencies

• scikit-learn
• numpy
• pandas
• matplotlib (for clustering visualization)

Tested with

• Direct script run: pass (demo returns valid JSON; classification and clustering scripts run end-to-end)
• Agno agent (Claude Haiku 4.5): pass

Agno agent verdict (excerpt)

The agent invoked the scikit-learn skill with a binding potency query, received a valid regression result with R-squared and scaling exponent, and confirmed the JSON output matched the expected schema for downstream pipeline integration.

Fix notes

• Fixed mutable range() default arguments in clustering_analysis.py (k_range parameter in two functions)
• Cleaned __pycache__/ directory
• Applied 8 ruff lint fixes