Skip to content

Dashboards in Neptune#

Example of a custom dashboard in Neptune

A dashboard is a customizable display option for runs in your project.

You can combine different metadata types, such as charts, parameters, image series, and code snapshots. For example:

  • Overlaid metrics and loss for training and validation
  • Model parameters, training code, hardware consumption, and ROC curves
  • Validation metrics and image predictions on a subset of a validation set next to each other

See example dashboard in Neptune 

Single-run dashboard#

If you navigate to Dashboards while inspecting a single run, available (and newly created) dashboards will be single-run dashboards. This means that the available widgets will be specific to analyzing one run at a time rather than contrasting multiple runs at once.

To create a new single-run dashboard:

  1. Click on a run.
  2. Navigate to Dashboards and select New dashboard.
  3. Enter a title for your dashboard.

  4. To add a widget, click New widget. You can either:

    • Type in the field you want to display and see which widget options you have, or
    • First choose a widget type, then enter fields to include.

    For an overview of widgets available for each metadata type, see Widgets and supported field types.

  5. Enter a widget title.

  6. Add more widgets to your dashboard as needed.
  7. Drag and resize the widgets to arrange them to your liking.

Multi-run dashboard#

If you navigate to Dashboards while having multiple runs selected, the available dashboards will be multi-run dashboards. This means that the available widgets will be specific to contrasting multiple runs at once.

For example, the scatter plot widget is available in this configuration.

To create a new multi-run dashboard:

  1. Toggle the eye icons ( ) to select multiple runs for comparison.
  2. Navigate to Dashboards and select New dashboard.
  3. Enter a title for your dashboard.

  4. Click New widget.

    1. Type in the field you want to display and see which widget options you have, or

    2. Choose a widget type directly, then enter fields to include.

      For an overview of widgets available for each metadata type, see Widgets and supported field types.

  5. Enter a widget title.

  6. Add more widgets to your dashboard as needed.
  7. Drag and resize the widgets to arrange them to your liking.

Scatter plots#

Scatterplot for accuracy and loss

If you have two metrics or parameters that you wish to compare or see how they relate to each other throughout the runs, you can create a scatter plot.

The scatterplot widget supports the following field types: Float, Int, and Datetime, in any combination.

To create a scatter plot:

  1. Ensure that multiple runs are selected (toggled on with the eye icon ).
  2. Create a new widget and choose the Scatter plot widget type.
  3. Click Add field and select a metadata field for each axis.
  4. Click Save widget.

You can edit the scatterplot later through the widget settings ( ).

Adding multiple charts at once#

Experimental

We're gradually introducing support for adding and displaying multiple metrics for multiple runs in the same widget.

Instead of manually creating a separate widget for each field you want to display, you can select multiple fields to be added at once.

In this case, the Create separate charts option is automatically selected, as showing multiple metrics on the same widget is currently only supported for a single run at a time.

To add multiple widgets at once:

  1. While having multiple runs selected ( ), create a new Chart widget.

  2. Enter a title that describes the metrics you're about to add. The name is duplicated for each widget ("WidgetName (0)", "WidgetName (1)", and so on).

    You can edit the widget titles later.

  3. Select the fields to be displayed.

    1. Click Select visible to check the selection boxes of all fields currently shown in the drop-down list.

      The preview only shows the first widget. The widges for the other selected fields are still created.

    2. To finish, click Add x selected and close.

  4. To create the widgets, click Add widget.

Exporting to report#

You can export a dashboard together with particular runs as a report. This way, you can share and return to the analysis of a particular set of runs later.

  1. Select Export to → Export to report.
  2. To confirm that you want to use the currently selected ( ) runs for the report, check the corresponding box.

On exporting, you're taken to the report automatically.

To access all reports, in the left nav, select Reports.

Versioning#

Dashboards are autosaved when you make a change.

You can access the version history through the button. In the version options ( ), you can name and restore particular versions of the dashboard.

Saving a dashboard with particular runs

A dashboard configuration doesn't take into account which runs are selected at the time.

To also save the run selection, create a report instead.

Widgets reference#

The following tables show which widgets you can use for each type of metadata when constructing dashboards.

Single run#

These widgets are available when inspecting a single run:

Metadata type Neptune field type Supported widgets
Single value (scores, text, timestamp) Float, Integer, Boolean, String, Datetime (logged with assign() or =) Single value
Numerical series (metrics, loss, accuracy) FloatSeries (int or float values logged with append() or extend())

Chart

Value list

Single value (for aggregates like variance, max, min)
Text series (journal, notes) StringSeries (str logged with append() or extend()) Value list
Namespace (parameters dictionary, monitoring namespace, system namespace) Namespace Parameters table
Single image File (image logged with upload()) Image
Series of images (e.g. predictions) FileSeries (images logged with append() or extend()) Image gallery
Single file (data sample, git diff, HTML) File (any file logged with upload()) File preview
Set of files FileSet (multiple files logged with upload_files()) FileSet preview
CSV file File (CSV logged with upload()) Table
Jupyter notebook checkpoint Notebook field generated by the neptune-notebooks extension (source_code/notebook field) Notebook preview
See supported fields per widget
Widget Neptune field type
Chart FloatSeries (1 or more fields)
Parameters table Namespace (folder)
Image Single image file
Image gallery FileSeries
Single value Float, Integer, Boolean, String, or Datetime
Value list StringSeries or FloatSeries
File preview File
FileSet preview FileSet
Table File containing CSV file
Notebook preview Notebook field generated by the neptune-notebooks extension

Multiple runs#

When multiple runs are visible ( ), the available widgets support displaying the same field from multiple runs at once.

Metadata type Neptune field type Supported widget
Numerical series (metrics, loss, accuracy) FloatSeries (int or float values logged with append() or extend())1 Chart
Numerical or datetime values Float, Integer, Datetime (logged with assign() or =) Scatter plot
Single image File (image logged with upload()) Single image comparison

Related

  1. Refers to a single metric. Combining multiple different metrics in a single chart is currently only supported for single-run widgets. ↩