skills.rest
hung-phan
hung-phanCOMMUNITY
Explore all skills by this author →
GitHub

online-learning

Detect concept drift and incrementally update models on streaming data.

1|Updated Jun 4, 2026
Search Tags:#drift detection#model monitoring#concept drift#online learning#incremental learning
One-click install
npx skills add https://github.com/hung-phan/ml-skills --skill online-learning
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: online-learning
Source: https://github.com/hung-phan/ml-skills/tree/main/skills/ml-review/references/ml-training/online-learning
Command: npx skills add https://github.com/hung-phan/ml-skills --skill online-learning

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires river, frouros, alibi-detect, sklearn, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This Skill helps you adapt models to streaming data, detect concept drift, and incrementally update models without full pipeline re-runs.

Core Features & Use Cases

  • Drift Detection: Detects covariate, label, and concept drift using PSI, KS test, ADWIN, DDM, and Page-Hinkley methods.
  • Incremental Learning: Supports partial_fit, warm-start, and true online learning with the river library.
  • Train/Score Separation: Implements the model-as-data control stream pattern for hot-swap without restart.
  • Event Time vs Processing Time: Handles event-time windowing and watermarks for accurate feature computation.
  • Feature Stores: Briefly frames the use of feature stores like Feast and Tecton for consistency and backfills.

Quick Start

Use the online-learning skill to detect drift in your model's input features and update the model incrementally.

Frequently Asked Questions about online-learning

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

FAQPage Schema
How do I detect concept drift in streaming data for machine learning models?▼

To detect concept drift in streaming data, you can use statistical methods like PSI, KS test, ADWIN, DDM, and Page-Hinkley to monitor covariate, label, and concept drift in real-time model features.

Can I incrementally update scikit-learn models without retraining the entire pipeline?▼

Yes, you can incrementally update scikit-learn models using the partial_fit and warm-start methods, allowing you to adapt to new streaming data without executing full pipeline re-runs.

How does incremental learning work with the river library for online model updates?▼

Incremental learning with the river library implements true online learning by continuously updating model weights as new streaming data arrives, supporting the model-as-data control stream pattern for hot-swapping models without restart.

Does this approach support event-time windowing and watermarks for feature computation?▼

Yes, it handles event-time windowing and watermarks to ensure accurate feature computation in streaming data pipelines, distinguishing between event time and processing time for reliable model monitoring.

What are the limitations of using alibi-detect and frouros for drift detection?▼

While alibi-detect and frouros provide robust drift detection, they require careful configuration of statistical thresholds for methods like ADWIN and Page-Hinkley to avoid false positives in highly volatile streaming data environments.

How do feature stores like Feast and Tecton integrate with online learning models?▼

Feature stores like Feast and Tecton frame the use of consistent feature computation and backfills, ensuring that online learning models receive accurate, time-aligned streaming data for incremental updates.

Related Skills

Explore More Data & Analytics Skills→
doanchienthangdevdoanchienthangdev
4

Feature Stores

Manage machine learning features with centralized definition, storage, and serving.

Community
Advanced
Patasse97Patasse97

senior-ml-engineer

Deploy production ML systems with monitoring, drift detection, and cost controls.

Community
Advanced
nahisahonahisaho
3

scientific-streaming-analytics

Enable real-time streaming analytics with River-based online learning, anomaly detection, and drift monitoring.

Community
Advanced
2nth-ai2nth-ai
1

MLOps

Deploy and maintain machine learning models in production with drift detection and retraining.

Official
Advanced