---

name: mlxtend description: "Use mlxtend machine-learning extension utilities for estimator ensembles, evaluation, feature workflows, frequent patterns, plotting, datasets, and small helper APIs." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

mlxtend

Use this repo skill when a task needs package-specific guidance for mlxtend (machine learning extensions): sklearn-style ensemble/meta-estimators, evaluation and model-comparison utilities, feature selection/extraction/preprocessing, frequent itemset and association-rule mining, Matplotlib diagnostics, packaged toy data, file/text/math helpers, or troubleshooting for those APIs.

Start here

1. Read references/package-overview.md for package scope, install/import checks, dependency notes, and cross-workflow routing.
2. Read references/troubleshooting.md for install/import, dependency, headless Matplotlib, sklearn compatibility, shape, slow-workflow, and file-group issues.
3. Read references/repo-provenance.md before deciding whether this skill is current for a different mlxtend checkout or package version.
4. Run scripts/check_mlxtend_env.py when you need to verify imports and, optionally, all bundled sub-skill smoke workflows.

Install and quick check

python -m pip install mlxtend
python - <<'PY'
import mlxtend
print(mlxtend.__version__)
from mlxtend.evaluate import accuracy_score
from mlxtend.frequent_patterns import apriori
print("mlxtend import ok")
PY

mlxtend is a Python API package; no first-class mlxtend CLI entry point is expected. For a deeper generated-skill check, run:

MPLBACKEND=Agg python scripts/check_mlxtend_env.py --run-subskill-smokes

Sub-skill routes

If the task asks about...	Read
Voting classifiers, stacking classifiers/regressors, classic mlxtend estimators, Kmeans, sklearn GridSearchCV parameter prefixes, predict_proba, sample weights, convergence, fit/predict behavior	sub-skills/estimators-and-ensembles/SKILL.md
Accuracy/scoring/confusion/lift, bootstrap/OOB/.632, permutation tests, paired t-tests, 5x2cv F tests, McNemar/Cochran tests, holdout splitters, grouped time-series splitters, feature importance, counterfactuals	sub-skills/evaluation-and-validation/SKILL.md
SequentialFeatureSelector, ExhaustiveFeatureSelector, ColumnSelector, PCA/LDA/RBF kernel PCA, scaling/standardization, sparse/dense transforms, one-hot labels, TransactionEncoder preprocessing	sub-skills/feature-workflows/SKILL.md
apriori, fpgrowth, fpmax, hmine, association_rules, transaction-to-one-hot schemas, rule metrics, empty itemset/rule diagnosis, large itemset memory choices	sub-skills/frequent-patterns/SKILL.md
plot_decision_regions, confusion-matrix/heatmap/learning-curve/SFS/PCA/scatter plots, headless Matplotlib, packaged datasets, find_files, find_filegroups, tokenizers, name helpers, math and utils	sub-skills/plotting-and-utilities/SKILL.md

Common route combinations

• Estimator analysis: use estimators-and-ensembles to build/fix the model, evaluation-and-validation to score or compare it, and plotting-and-utilities to visualize decision regions, confusion matrices, or learning curves.
• Feature engineering: use feature-workflows for selectors/transforms, then route to estimators-and-ensembles or evaluation-and-validation for downstream model checks.
• Market-basket mining: use feature-workflows for TransactionEncoder details when needed, then use frequent-patterns for itemsets/rules and threshold troubleshooting.
• Headless examples or CI checks: set MPLBACKEND=Agg, run the nearest sub-skill smoke script, and close/save Matplotlib figures rather than displaying them.

Boundaries

• Do not use this skill for generic scikit-learn, pandas, NumPy, or Matplotlib tasks unless the user specifically needs an mlxtend API, example, failure mode, or compatibility detail.
• Do not use this skill for deep-learning training/serving, LLM workflows, distributed compute, experiment tracking, or repository-maintenance tasks unrelated to mlxtend's public package APIs.
• All selected workflows are CPU Python APIs. No CUDA, ROCm, MPS, external service, credential, or model-download backend is required by this generated skill.