An autonomous machine learning agent that iteratively evolves models and leverages multi-layer memory to dominate Kaggle competitions and tabular data tasks.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install ml-evolution-agent
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install ml-evolution-agent using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is ML Evolution Agent?
The ML Evolution Agent is a sophisticated tool designed for the Openclaw Skills ecosystem, specifically built to handle the complexities of competitive machine learning. Inspired by the MLE-Bench methodology, it functions as a self-improving system that tracks every experiment to refine its approach over time. By utilizing a unique Hierarchical Competency Cluster (HCC) memory architecture, it bridges the gap between raw data processing and strategic decision-making.
This skill is particularly powerful for developers and data scientists who need to scale their experimentation without manual intervention. It doesn't just run scripts; it learns which feature engineering techniques provide the highest gain and which model configurations are most stable under specific resource constraints. As one of the most advanced Openclaw Skills, it ensures that your ML pipeline is always moving toward a higher Leaderboard (LB) score through automated experimentation.
ML Evolution Agent Use Cases
- Automating end-to-end Kaggle competition pipelines to reach specific target leaderboard scores.
- Solving complex tabular data classification and regression tasks with minimal manual intervention.
- Conducting large-scale automated ML experimentation where iterative learning and memory are required.
- Managing resource-intensive ML tasks that must respect strict system limits like API quotas, time, and memory.
How ML Evolution Agent Works
- The agent initializes the competition environment and sets a target Leaderboard score based on user requirements.
- It pulls baseline data and applies initial feature engineering patterns stored in its Knowledge Memory layer.
- The evolution loop begins, where the agent generates and executes training scripts using models like CatBoost, XGBoost, or LightGBM.
- After each phase, it evaluates Cross-Validation (CV) and Leaderboard (LB) scores to determine if the experiment was a success.
- It updates its HCC Multi-layer Memory, recording episodic logs and identifying new success patterns for the next iteration.
- The agent applies strategic rules to decide whether to roll back to a previous state, fine-tune parameters, or explore new feature directions.
ML Evolution Agent Setup
To get started with this powerful addition to your Openclaw Skills, ensure you have Python 3 and the Kaggle CLI installed. You can then install the agent using the following command:
clawhub install ml-evolution-agent
Once installed, initialize the evolution agent in your Python script to start the training process:
from ml_evolution import MLEvolutionAgent
agent = MLEvolutionAgent(
competition="your-competition-name",
target_lb=0.95400,
data_dir="./data"
)
agent.evolve(max_phases=10)
ML Evolution Agent Data Schema & Taxonomy
The ML Evolution Agent organizes its workspace into a structured directory to maintain its evolving state and library of techniques:
| File/Folder | Description |
|---|---|
HCC_MEMORY.md |
Detailed architecture of episodic, pattern, knowledge, and strategic memory layers. |
FEATURE_ENGINEERING.md |
A dynamic library of feature techniques (like Target Statistics) and their recorded impacts. |
MODEL_CONFIGS.md |
Optimal hyperparameter configurations for major gradient boosting frameworks. |
EVOLUTION_RULES.md |
The logic tree for auto-evolution decisions and overfitting detection protocols. |
templates/ |
Python scripts used for training both baseline and evolved model versions. |
memory.json |
The machine-readable state file representing the agent's current knowledge and experiment history. |
ML Evolution Agent Advanced Features
- Multi-layer HCC Memory architecture that persists learning across different competition phases and sessions.
- Automated decision trees that handle rollbacks and direction changes based on performance degradation metrics.
- Integrated overfitting detection that automatically increases regularization or simplifies models when CV/LB gaps widen.
- Resource-aware execution logic that prioritizes feature selection to stay within strict training time and memory limits.
- Native support for weighted ensembles, utilizing optimal configurations for CatBoost, XGBoost, and LightGBM.
SKILL.md
Loading
Related Openclaw Skills
Memory Sync Enhanced
A dual-layer AI memory architecture that mimics human biological forgetting and associative learning patterns to optimize context relevance.
Memory Sync System (CortexGraph)
An automated memory synchronization system that bridges local Markdown files and CortexGraph using intelligent forgetting curves.
MBTI Agent - Digital Sprite Personality Assessment
A specialized psychometric framework for evaluating the personality profiles of non-interactive AI agents using a static 70-question assessment.
Market Analysis Service
A data-driven analytical suite providing industry trends, competitor deep dives, and user behavior insights.
MLOps Automation
A comprehensive automation suite for standardizing machine learning tasks, containerization, and continuous integration pipelines.
MLOps Collaboration
A professional toolkit for preparing MLOps projects for seamless community collaboration and open-source distribution.
Featured*
