Displaying metadata

You can display, explore, and compare, model building metadata for every Run you log to Neptune.

To find the metadata you logged to the Neptune UI:

See example single Run UI in Neptune

Displayed model training metadata for a single Neptune Run

Neptune UI for a single Run shows:

  • All metadata: all model building metadata logged to a Run presented in a directory structure.

  • Charts: all numeric series (metrics, losses) visualized as charts.

  • Monitoring: information about the utilization of the memory, CPU, and GPU, console logs (stdout, stderr).

  • Source code: code snapshots, git information.

  • Custom dashboards: a combination of various metadata types (metrics, code, charts, images) presented in one place.

  • Details: additional Run information like tags, ID, who and when created a Run.

All metadata

Model-building metadata is presented in a directory structure or namespaces inside of the All metadata section of the single Run UI. It keeps the exact structure in which you logged it.

See an example All metadata section in the Neptune Run

Directory structure of namepsaces in the Neptune UI

Note Neptune creates source_code, monitoring, and sys namespaces (directories) in All metadata. Those are automatically displayed in Source code, Monitoring, and Details sections of the UI.

To see metadata displayed for a particular Field, click through the directory structure and find it.

Neptune displays:

  • images and image series,

  • metrics and learning curves,

  • interactive visualizations,

  • hardware consumption,

  • model parameters,

  • and more.

To see a full list of supported metadata types, see What can you log and display.

See an example image series displayed in the Neptune Run

Image metadata displayed in Neptune

Charts

You can see all series of numeric values, like metrics and losses during training, visualized as charts in the Charts section of the single Run UI.

You can filter your charts by typing a string into the Filter charts box.

See an example Charts section in the Neptune Run

Note If you want to overlay multiple metrics on one chart or create a dashboard that only shows some selected metrics, read the section on Creating custom dashboards.

Monitoring

You can log hardware consumption and console logs (stderr, stdout) to Neptune whenever you create a Run and see it in the Monitoring section.

Specifically, you can see:

  • cpu: total CPU consumption of your machine,

  • memory: total memory consumption of your machine,

  • gpu: GPU consumption of your machine,

  • gpu_memory: GPU memory consumption of your machine,

  • stderr: standard error logs from your console,

  • stdout: standard output logs from your console.

See an example Monitoring section in the Neptune Run

Hardware consumption logged to Neptune

Source code

You can log code to Neptune whenever you create a Run and see it in the Source code section.

See an example Source code section in the Neptune Run

Displaying code snaphot logged to Neptune

Important You can turn off logging of source code by passing an empty list to the source_files argument of neptune.init():

run = neptune.init(..., source_files=[])

If you have a .git directory inside your project, Neptune will automatically log some information from your commit history.

Displaying git information in Neptune

Creating dashboards

You can combine different metadata types like charts, parameters, image series, code snapshots, and more in a single place by creating dashboards.

Dashboards are templates that let you customize what you want to see for any Run in a project. For example, they let you:

  • See overlayed metrics and loss for training and validation,

  • See model parameters, training code, hardware consumption, and ROC curves in a single place,

  • See validation metrics, and image predictions on a subset of a validation set next to each other.

Important Dashboards are created inside of a single Run UI but are visible in all Runs in the Neptune project.

See an example Dashboard in a the Neptune Run

Example dashboard in Neptune

How to create a custom dashboard

To create a custom dashboard you need to:

  • Click on + Add new dashboard.

Finding a button to Add new dashboard in Neptune.
  • Choose a title, color, and icon for your dashboard.

Choosing name, color and icon for the Neptune dashboard.
  • Click + Add widget button at the top right.

Finding +Add widget in the Neptune UI.
  • Choose a widget type and click Add widget. You can display charts, parameters, images, code, and more.

Choosing the Paramters table widget.
  • Choose a widget title and add metadata to your widget.

Choosing model parameters for the Parameters table widget.
  • Save your new dashboard with a first widget by clicking Save.

Model parameters displayed in a Neptune dashboard.
  • Add new widgets to your dashboard.

Showing multiple metrics on one chart

If you want to put multiple metrics on one chart, for example, to see a learning curve for a training/validation loss, add a Chart widget to your dashboard.

To do that you need to:

  • Go to your custom dashboard or create a new dashboard.

  • Click +Add new widget.

  • Choose a Chart widget and click Add widget.

Choosing Chart widget.
  • Type in a metric name and add it by clicking.

Finding metrics for the Chart widget.
  • See if all the metrics were added properly and are visible in the Fields to be displayed on the widget.

Choosing metrics fot the Chart widget.
  • Click Add widget and see your learning curves added to your dashboard.

Example of two overlayed metrics on one chart in Neptune.
  • Save changed to the dashboard or add more widgets.

Combining many metadata types on one dashboard

If you want to see various metadata like images, model parameters, and learning curves, add various widget types to your dashboard.

  • Go to your custom dashboard or create a new dashboard.

  • Click +Add new widget.

  • Add learning curves by using a Chart widget. Click Add widget and choose a Chart.

Choosing Chart widget.
  • Add your metrics to a Chart and click Add widget.

Adding metric fields to the Chart widget.
  • Add model parameters by using a Parameters table widget. Click Add widget and choose a Parameters table.

Choosing Parameters table widget.
  • Choose a namespace that you would like to present as a parameters table.

Adding model parameters fields to the Parameters table widget.
  • Add image data sample by using an Image gallery widget. Click Add widget and choose an Image gallery.

Choosing an Image gallery widget.
  • Choose a Field with a list of images logged to it.

Adding image fields to an image gallery widget.
  • See learning curves, model parameters, and a sample of the image data on a single dashboard.

Dashboard with learning curves, model parameters and images in Neptune.
  • Save your dashboard or add more widgets.

Details

You can find additional information, about the Run in the details section. Specifically:

  • ID: id of a Run that was created automatically

  • State: whether the Run is active or inactive

  • Name: name you specified for the Run

  • Tags: tags you appended to the Run. You can also add tags directly from the UI.

  • Description: a field where you can add comments about the Run, even directly from the UI

  • Owned by: who created the Run

  • Created: when the Run was created

  • Modified: when the Run was last modified

To see or modify the Run details:

  • Click on the Details in the single Run UI. It should open a modal window.

Where to find Details section of the single Run UI
  • See or modify Run details

Where to see and how to modify Run details

What's next?