A flexible UI for exploring model behavior.

The Gantry Dashboard

The dashboard summary page provides an overview of metrics, events, and triggered alerts by application.


The application page lists all the instrumented ML-driven applications configured with Gantry.

Associated with each application is a chart that denotes the quantity of data ingested along with the timestamp.


NOTE: The default data time window that is retrieved is set for 1 day across all applications.

Let’s get more insights into the data that you’ve passed to Gantry! Click on an Application to see more.


A application overview page for a new application where data has not yet been processed.

The page for a new application may look like the one above. You can add metrics easily. Below, you can find the overview page for an application with favorite metrics configured and triggered alerts.


Overview page for a sample production application.

The application overview page shows your favorited metrics and summaries of triggered, open alerts. This page has metrics already configured. You can add them easily from here by scrolling down to "add favorite metrics".

You can click on the Ingestion panel in the upper left to go to the jobs page to check on batches and backfills.

Now, click on the Explore button on the left sidebar, just below the currently-selected overview button, to explore your data.

The Explore page


The Explore page has three main areas.

The top bar is for filtering, comparing, and grouping the data in your application: creating views.

The middle area is a timeline of the records in your application over the selected time interval, as filtered by the current view.

The bottom area has three tabs: Data, Distributions, and Timeline.

We'll talk about those three areas, first, before we talk about creating views of data.


The Explore page.


The smart data viewer shows you the records in the currently filtered view of your application's data. You can click or brush in the histograms of the data viewer's headers to quickly filter by particular values.


We'll show you how to set up more complex filters in the creating views section, below.

On the right hand side of the Data tab, you'll find controls that allow you to easily configure columns in the data table and download your data for analysis.


The show/hide button and the download pulldown.


You can view the the distributions of your data, grouped by inputs, outputs, feedback, and projections, in this tab. It also has a show/hide selector which allows you to view only the fields you want to focus on.


The Distributions tab.


The timeline tab is where you can create and configure metrics to help you understand model performance. For a new application, it may have some metrics that Gantry has inferred may be useful due to your input, output, feedback, and prediction datatypes.

You can create your own metrics by clicking on the "New chart" button on the right and following the prompts.


Creating a new chart in the Timeline tab

Creating views

Along the top of the Explore page is is the view creation bar. You can filter, group, and compare your data with views.

First, select the version of your application you wish to examine. You can then select filter fields, time interval, and a way to group the data, as shown below.


Filtering data for the last 12 hours, only including divorced or single people, and grouping by the amount of their loan. Gantry automatically figures the grouping tranches for you, based on the data.

Saving and comparing views

Now we can save this view and make it a baseline for comparison.


Saving a view saves the time range, the filter, and the grouping information for later viewing.

Clicking "Compare to a baseline" will shift the currently active view to the baseline and allow you to set the filters easily for comparison.


You can make the current view a baseline for comparison.

What’s Next