Reproducibility is a challenging problem in Machine Learning, it’s the concept of being able to recreate a machine learning workflow or experiment and reach the same results.

Polyaxon makes your experiments reproducible, portable, and repeatable while being language and framework agnostic.

Packaging Format

Every operation in Polyaxon is authored using a powerful specification and packaging format Polyaxonfile.

Polyaxonfile is a specification for packaging dependencies, inputs, outputs, artifacts, environments, and runtime of an operation to schedule on Kubernetes.

Please refer to Polyaxonfile specification for more details.

Tracking

Polyaxon comes with a built-in extensive tracking system. You can log information for source code, parameters, data, metrics, tags, and logs using Polyaxon APIs or clients.

Please refer to the tracking API for more details.

Lineage

Every operation scheduled with Polyaxon will be auto-documented with lineage information about its inputs and outputs, statuses, metrics, hyperparams, source code, data, visualizations, artifacts, and resources used in each experiment.

Polyaxon provides a dashboard and a CLI to see the full history at a glance, including when, who, and where, as well as an advanced insight and comparison of experiments based on results, hyperparams, versions of training data, and source code.

Please refer to the runs dashboard for more details.