Polyaxon comes with a powerful and customizable visualization system for driving insights in the dashboard or programmatically.

run-dashboards

Native visualization system

Polyaxon dashboard provides several visualizations and charts based on events logged by the tracking API.

For single runs:

  • Line charts
  • Bar charts
  • Stats charts
  • Histogram charts
  • Curve charts: PR curves, AUC/ROC curves, custom curves (<x, y> arrays)

For multi-runs you can compare metrics and curves from several runs, as well as charts for hyperparameter values and metrics:

  • Line charts
  • Bar charts
  • Stats charts
  • Curve charts: PR curves, AUC/ROC curves, custom curves (<x, y> arrays)
  • Scatter charts
  • Parallel coordinates
  • Contour plots

Media rendering

Polyaxon can render several media types, you can log several versions of the same media with a step index:

  • images
  • text
  • html
  • audio
  • video
  • dataframe

Custom Charts

Oftentimes users have advance use cases that are not supported by the platform, or they need to integrate visualizations generated by external systems.

Polyaxon can consume any chart specification generated by these libraries:

  • Matplotlib
  • Plotly
  • Bokeh
  • Altair/Vega

Programmatic Experience

Polyaxon has a programmatic experience for generating visualizations for single runs or multi-runs using:

  • Plotly express
  • HiPlot

Tensorboard

Tensorboard is a tool that is used by data scientists to visualize and debug deep learning models. Polyaxon provides several ways for using tensorboard:

  • Tensorboard for a single run
  • Tensorboard for a multi-runs
  • Tensorboard based on metric(s) performance or a search

Example

You can run the artifacts example which calls several tracking methods and logs several metadata and artifact types.