Experiment Tracking

Experiment tracking is the process of saving all experiment related information that you care about for every experiment you run.

The basics

Machine learning models are an outcome of dozens and hundreds of experiments to discover the correct parameters, algorithm and data. In this process it is important to track changes that worked and the one's that did not.

Experiment tracking is designed for ML practitioners to help make sense of these iterations, compare and learn from each that would enable them to make systematic progress to improve the next iteration.

Newron experiment tracking lets you log source properties, parameters, metrics, tags, models and artifacts related to training a machine learning model.

A Newron project is used to co-ordinate the ML development towardas an objective like "Product Recommendation" for an e-commerce site. To get there the team may try multiple approaches, it could be a deep learning based approach or some algorithms from classical machine learning. Each of these approaches would be an experiment. For each experiment you would do iterations to tune the particular algorithm. The iterations within an experiment is called runs.

An Experiment is the primary unit of organization and access control for runs; all runs belong to an experiment. Experiments let you visualize, search for, and compare runs, as well as download run artifacts and metadata for analysis in other tools. You can compare experiments within a project to compare high level metrics to determine the best approach for the given problem.