ml-experiment

To track ML experiment metrics and compare runs, this skill logs hyperparameters and key metrics to an experiment registry, generating comparison plots to analyze accuracy, loss, and runtime across multiple runs for quick decision-making.

To make PyTorch Lightning training runs reproducible, this skill saves environment package versions, git commit hashes, and resolved Hydra configurations, allowing you to accurately rerun any stored experiment setup later.

Yes, the ML experiment tracking skill supports logging and metric-based selection using W&B, TensorBoard, or MLflow, integrating these platforms into your training workflow to manage experiment registry logging.

You can generate Hydra configs for ML experiment sweeps and ablations by using this skill to standardize baseline configurations, creating structured files that standardize what you run before executing training.

The best way to compare ML hyperparameters across multiple training runs is to record them into an experiment registry JSON file, using companion scripts to generate comparison plots and select models based on metrics.

You need to log experiment status to a JSON registry file to centralize configs and metrics in a single workflow, preventing the loss of insight and time when managing many machine learning experiments.