Act and Share

Now that you've reviewed DataRobot Insights for the recommended model, let’s take a look at a few ways that you can act on and share your work—starting with one of my favorite examples of automation in DataRobot.

Navigate to the Leaderboard and you’ll see the link for Download Report. This automatically generated, fully customized report summarizes what you did in DataRobot to build your model. It provides an overview of how many models it trained, how it evaluated those models, why it made its recommendation, and how the features in the dataset impacted the prediction. It's in a form that you can revise and edit for use with your stakeholders, your colleagues, and even regulators to defend your process and build transparency about what’s driving the model. Imagine how much time this alone will save you!

lhaviland_1-1601576965038.png

Here’s an example of what the first page of an AI Report looks like:

lhaviland_2-1601577017363.png

You can also download the exportable charts created by your modeling exercise and use these as another basis for delivering your results and recommendations to your stakeholders. With the model selected in the Leaderboard, navigate to Predict > Downloads tab. 

lhaviland_0-1601920617202.png

The content available for download includes images and data related to hot spots, feature effects, feature impact, and other facets of the analysis. You can use these assets along with the AI Report when you prepare your presentation to your stakeholders about ways to reduce the frequency of late shipments and the data-driven reasons behind your recommendations.

Another way to act on your model is to deploy it (from the Leaderboard, select the recommended model and navigate to the PredictDeploy tab). Creating a deployment allows you to leverage your model against new data about upcoming shipments to determine if they might be late. 

 

lhaviland_0-1601917202018.png

In the Applications section of DataRobot, you can use the deployed model to assess potential outcomes of various scenarios that you might consider in your proposals to reduce late shipments.

Finally, let’s navigate to the Applications tab. Apps are a very easy way to share your work with your colleagues and test your theories on how to make better decisions.

We saw earlier that shipments which originate at a Regional Distribution Center are often late. But, as compared to a direct delivery, how late? We can use the Predictor app to get this answer (Applications tab > Predictor).

lhaviland_0-1601917412695.png

As shown in the Predictor app, direct drops have a 39% probability of being late, whereas shipments from distribution centers have an 85% probability of being late. You can use the Predictor app to create a variety of different scenarios to evaluate possibilities for changes to improve shipment times.

 

Direct drops—% probability of being lateDirect drops—% probability of being late

From distribution centers—% probability of being lateFrom distribution centers—% probability of being late

More information

Predictor App Overview
(Also, see the introduction to all articles in this series) 


This completes our business analyst's journey through DataRobot. Please comment to this article with any questions you may have or any feedback you'd like to share, and don’t forget to check out the DataRobot Community for more resources and dialogue (starting with the business analyst index). We look forward to hearing from you!

Comments
Blue LED

@KarinAISD- I've been looking for an explanation like this. Thanks for sharing. I'll be sending it along to my team

Version history
Revision #:
15 of 15
Last update:
3 weeks ago
Updated by:
 
Contributors