sacred

Why we included this project

Running many machine learning experiments often means losing track of which configuration produced which result. Sacred wraps your training scripts so each run is automatically logged with its configuration and metrics, plus any artifacts, and it can store that history in MongoDB for later querying. The configuration system lets you define defaults and override them from the command line, which makes sweeping hyperparameters and reproducing a specific run far less error-prone. Teams that value clean experiment records will find the observer-based logging and the ability to compare runs useful without having to adopt a heavyweight platform. It is a practical, minimal-overhead choice for researchers and small teams who want reproducible experiments without restructuring their existing code.