ml-meta-labeler

Automate ML meta-labeling for trading with triple-barrier labeling and XGBoost calibration.

Updated Apr 11, 2026

Search Tags:#calibration #trading #xgboost #meta-labeling #triple-barrier #purged-cv

Ready to run

npx skills add https://github.com/bitandbytes/Argus --skill ml-meta-labeler

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: ml-meta-labeler
Source: https://github.com/bitandbytes/Argus/tree/main/.claude/skills/ml-meta-labeler
Command: npx skills add https://github.com/bitandbytes/Argus --skill ml-meta-labeler

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This skill automates the meta-labeling layer (Layer 3) of a multi-model trading cascade, enabling selective action on quant-engine signals to improve precision and reduce false positives.

Core Features & Use Cases

Triple-barrier labeling to generate robust binary/meta labels for training.
XGBoost calibration and Platt scaling to produce well-calibrated trade-probabilities.
Purged K-fold CV with embargo to prevent data leakage and maintain realistic out-of-sample evaluation.
Threshold-based decisioning and routine retraining to adapt to changing markets.

Quick Start

Provide a history of quant signals and features, then run the meta-model training workflow to produce a calibrated classifier.

Frequently Asked Questions about ml-meta-labeler

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How does meta-labeling improve trading signal precision?▼

Triple-barrier labeling generates robust binary or meta labels for training by setting profit-taking, stop-loss, and time barriers around each trade entry. This mechanism captures the actual outcome dynamics of trading signals to produce realistic target variables.

How do I train an XGBoost meta-model using purged cross-validation?▼

To train an XGBoost meta-model, assemble a feature matrix from quant and regime inputs, then apply embargo-based purged K-fold cross-validation. This prevents data leakage during training and maintains realistic out-of-sample evaluation for robust production models.

Why use CalibratedClassifierCV and Platt scaling for trade probability calibration?▼

You must provide a historical dataset of quant signals and a feature matrix assembled from quant features, market regime inputs, and sentiment data. This historical signal and feature history is required to run the meta-model training workflow and produce a calibrated classifier.

What are the limitations of using purged cross-validation with embargo for trading models?▼

Purged cross-validation with embargo prevents data leakage by removing correlated data points around training folds, which limits the effective training dataset size. This approach requires sufficient historical data to maintain statistical significance for robust out-of-sample evaluation.