Comment on page
Tensorleap provides real-time and historical metrics monitoring.
The metrics defined in the Integration Script are stored during model Evaluation / Training for each sample, batch, and epoch. In addition, each sample is monitored and stored with its metadata.
All stored metrics and metadata can be presented aggregatively in a custom Dashlet.
A Dashboard is a panel that layouts various metric charts, also called Dashlets.
Dashboards in Tensorleap are presented via the Dashboard panel, which is shown when you click the
tab at the top.
To add a new Dashboard, click
at the top bar.
The dashboard displays visualizations of the currently selected model(s). To select models, click
to open the Versions panel on the left, expand the version, and select the model(s).
The dashboard will automatically update and show data for the selected models. At the top bar, you can see tags with the selected models. From there, you can also toggle to filter models.
Models Selection for the Dashboard
A Dashlet is an interactive visualization that updates in real-time, displaying aggregated metrics against various parameters. To add a new Dashlet click
and choose the chart type on the left.
The different Dashlet types, and their properties, are described below.
F1 metric Dashlet
Balanced Accuracy Dashlet
PR Curve Dashlet
ROC Curve Dashlet