Can I customize the drawdown thresholds?

Yes, the skill provides standardized NativePPOConfig defaults that can be adjusted for different contexts, such as training validation vs. live trading.

Why was my training stopping at 1000% drawdown?

This typically occurs when drawdown is calculated as an absolute value rather than a percentage of the running peak; this skill implements proper percentage-based logic.

What is adaptive recovery in ML training?

Instead of a hard stop, adaptive recovery reduces the learning rate and increases entropy to encourage exploration, allowing the model to recover from performance dips.

How does this skill handle PPO reward signals?

It provides a workflow to convert PPO reward signals into a simulated equity curve, ensuring that stop-loss triggers are based on meaningful relative performance.

Training Resilience for ML Trading

Name: Training Resilience for ML Trading
Author: smith6jt-cop

bysmith6jt-cop

0•

Data Science & ML

Optimizes Reinforcement Learning training by preventing premature stops and implementing adaptive recovery for trading models.

The Training Resilience skill enhances Claude's ability to debug and optimize Reinforcement Learning (RL) training pipelines, specifically focusing on PPO-based trading models. It addresses common pitfalls such as incorrect drawdown calculations (absolute vs. percentage), inappropriate early-stop triggers, and the confusion between PPO reward signals and actual equity curves. By implementing adaptive recovery mechanisms—including automated learning rate reduction and entropy adjustments—it ensures training sessions are resilient to transient volatility and provides a robust framework for developing stable algorithmic trading agents.

Key Features

01Provides guardrails to prevent training crashes on false-positive spikes

020 GitHub stars

03Corrects percentage-based drawdown calculations for RL rewards

04Configures resilient early-stop thresholds tailored for PPO training

05Clamps metric ranges to ensure valid mathematical inputs for optimizers

06Implements adaptive recovery logic with LR and entropy adjustments

Use Cases

01Automating hyperparameter adjustments when models hit performance plateaus

02Standardizing validation metrics for PPO-based algorithmic trading agents

03Debugging impossible drawdown values in RL training logs

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add smith6jt-cop/skills_registry training-resilience

For use in Claude.ai and ChatGPT

Download Skill