Optimize black-box parameters using Nevergrad evolutionary algorithms.
npx skills add https://github.com/pengqianhan/scicomp-skills --skill nevergrad-skill
Please help me install this Agent Skill. Skill: nevergrad-skill Source: https://github.com/pengqianhan/scicomp-skills/tree/main/nevergrad-skill Command: npx skills add https://github.com/pengqianhan/scicomp-skills --skill nevergrad-skill
This skill solves the challenge of optimizing parameters for black-box systems, simulations, or machine learning models where gradients are unavailable or the objective function is non-differentiable.
Use the nevergrad skill to optimize the parameters of my simulation function by running the setup script and then calling the optimizer minimize method.
High-intent search queries and answers about installing and using this skill.
To optimize parameters for a simulation without gradients, you can use gradient-free black-box optimization. This approach accesses over 100 derivative-free optimizers to efficiently search complex continuous or discrete parameter spaces.
Hyperparameter tuning without gradients is best handled by evolutionary algorithms. They efficiently search for optimal learning rates or batch sizes by evaluating derivative-free objective functions across multiple CPU cores for parallel evaluation.
Yes, black-box optimization supports mixed parameter spaces for non-differentiable objective functions. The Nevergrad library provides specific optimizers designed to handle continuous, discrete, and mixed variables seamlessly.
To run derivative-free search strategies, set up a Python environment with Nevergrad and NumPy dependencies installed. Execute the setup script provided, then call the optimizer minimize method on your objective function.
Yes, you can scale evolutionary algorithm optimization tasks across multiple compute cores. The parallel evaluation feature distributes objective function assessments, significantly reducing total compute time for complex simulations.
You should avoid using gradient-free optimization when your objective function is fully differentiable and gradients are easily computable. In those cases, gradient-based methods typically converge faster than derivative-free search strategies.
cptnfrenCollect Linux host hardware and runtime context into a Markdown baseline report.
doanchienthangdevImplement machine learning training pipelines with PyTorch, Optuna, and MLflow.
gracefullightGenerate static, animated, and interactive plots with Matplotlib.
snoodleboot-ioExplain hyperparameter tuning techniques including grid search, random search, Bayesian optimization, and reinforcement learning.
