skills.rest
mzqef
mzqefCOMMUNITY
Explore all skills by this author →
GitHub

hyperparameter-optimization

Automate unified PPO hyperparameter and reward-weight optimization for robotic navigation.

Updated Feb 6, 2026
Search Tags:#PPO#reinforcement-learning#hyperparameter-tuning#bayesian#AutoML#grid-search#reward-weights
One-click install
npx skills add https://github.com/mzqef/MotrixLab --skill hyperparameter-optimization
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: hyperparameter-optimization
Source: https://github.com/mzqef/MotrixLab/tree/main/.github/skills/hyperparameter-optimization
Command: npx skills add https://github.com/mzqef/MotrixLab --skill hyperparameter-optimization

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill automates the joint tuning of PPO hyperparameters and reward/penalty weights, enabling a single, automated search to improve training efficiency, stability, and policy performance in robotic navigation tasks.

Core Features & Use Cases

  • Unified AutoML for PPO parameters (learning rate, entropy, clipping, epochs) and reward scales
  • Supports grid, random, and Bayesian search strategies with constraint-based filtering to skip unstable configurations
  • Provides a ready-to-run Quick Start and analysis tooling to compare configurations and export best results

Quick Start

uv run starter_kit_schedule/scripts/automl.py --mode stage --budget-hours 12 --hp-trials 8 Get progress with uv run starter_kit_schedule/progress/automl_state.yaml

Frequently Asked Questions about hyperparameter-optimization

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

FAQPage Schema
How do I tune PPO hyperparameters and reward weights at the same time?▼

You can jointly tune PPO hyperparameters and reward weights by defining a unified search space that covers learning rate, entropy, clipping, epochs, and reward scales, then running an automated search pipeline to evaluate configurations.

What is the best way to automate reward weight optimization for robotic navigation?▼

Automated reward weight optimization for robotic navigation applies grid, random, or Bayesian search strategies across reward scales and PPO parameters to improve training convergence, stability, and policy performance.

Can I skip unstable PPO configurations during hyperparameter tuning?▼

Yes, you can skip unstable PPO configurations by applying constraint-based filtering in your search pipeline, which automatically prunes configurations that violate stability constraints before full evaluation.

Does Bayesian search work for PPO hyperparameter tuning in reinforcement learning?▼

Bayesian search works for PPO hyperparameter tuning by efficiently exploring the joint space of PPO parameters and reward weights to find optimal configurations within a set budget.

How do I start an automated PPO hyperparameter search with a time budget?▼

You start an automated PPO hyperparameter search by running the automl script with a specified time budget and number of trials, then tracking progress through the generated state YAML file.

Why does my PPO training lack stability during reinforcement learning experiments?▼

PPO training lacks stability when hyperparameters and reward weights are misconfigured, but applying a coordinated search pipeline with constraint-based filtering prunes unstable configurations and improves convergence.

Related Skills

Explore More Software Engineering Skills→
mzqefmzqef

curriculum-learning

Orchestrate multi-stage curriculum training for robotic navigation tasks.

Community
Advanced
MesferAliMesferAli

openrlhf-training

Train LLMs with PPO, GRPO, RLOO, and DPO using Ray and vLLM.

Community
Advanced
softmgsoftmg

AgentDB Learning Plugins

Create and train reinforcement learning plugins with AgentDB and agentic-flow.

Official
Advanced
snoodleboot-iosnoodleboot-io
2

hyperparameter-optimization

Explain hyperparameter tuning techniques including grid search, random search, Bayesian optimization, and reinforcement learning.

Official
Intermediate