Getting SaaSsy with DataRobot

Dalton_Ruer · Sep 8, 2022 8:51:25 AM

With Qlik Sense Enterprise on Windows you needed to utilize a Server Side Extension to reach your data science platforms. But with Qlik Cloud you have a myriad of built-in advanced analytics/data science connectors at your fingertips. I hope that you will forgive me for the cheesy pun title, but it was hard to resist. As this document is about how you can call your DataRobot deployments from your Qlik Cloud SaaS tenant applications.

In this document, you will learn how to: Create an Analytics Connector to make predictions via DataRobot, Configure the connection and Call it to make your predictions.

Creating a Connection

The DataRobot connector will be found under the Analytics Sources when you choose to add a connection. Simply click it. Seriously it's that easy and probably didn't even need its own section. But Creating, Configuring and Calling seemed like a trifecta of headers so I went with it.

Configuring your Connection

Your first configuration option will be regarding the type of predictions you wish to make. Which of course are based on the deployment you are calling within DataRobot. The default is Predictions, but you can also choose Timeseries Predictions if that is the model you are calling from DataRobot.

The next several parameters involve things that are not simple drop-down selections like that. The Deployment ID, API Key and DataRobot Key are all really weird, unique number strings. The kind of things you can't usually memorize or rattle off the top of your head. The Host is a URL string that is unique to your deployment environment within the DataRobot universe. But don't panic they can be found pretty easily within your DataRobot SaaS environment. Since go to the "Deployments", choose the specific "Deployment" you wish to call, Choose "Predictions", "Prediction API" and then choose "Real-Time" as the Prediction Type.

As the following screenshot depicts, DataRobot will hand you all of that information.

Simply copy/paste the information provided by DataRobot to the "Create new connection (DataRobot)" dialog.

Notice that since the connection is capturing the "Deployment ID," that your DataRobot connections are specific to individual deployments. If a deployment is removed and a model is re-deployed, you will need to update your DataRobot connection within Qlik Cloud.

There are a few more parameters that you will need to populate. The "Timestamp Format" will be auto-filled for you, as will the "Response Table." If you are only calling a single DataRobot deployment within your application a name like "DataRobot Predictions" is fine. However, if you want to be super specific, or you will have multiple DataRobot deployments being called, you can rename it as desired.

Perhaps most importantly, is the Association section. As you know, tables within Qlik Sense are associated. So we need to provide the connector with the name of the KEY field/Association Field that is in our Data Model and check the box for "Send Association Field" so that when the data is returned (see below) it can be associated.

Like any of the data connectors within Qlik Sense, you need to provide a Name for your DataRobot connection. Then you are ready to Test the connection by pressing "Test connection" and then press "Create." (The following screenshot was taken after an Edit, so it has "Save" instead of "Create.")

Calling your Connection

Hurray! You have a new connection ready to go. Just like all data connection types, the DataRobot connection you just created needs to be called to retrieve data. Simply find your specific connection and click the "Select data" option.

Next you will be prompted for a really important piece of information, the name of the table that contains the data you need to have the predictions executed for. If you are a Qlikkie then you are well aware of the concept of a "Resident Table." If you aren't a Qlikkie, this is simply the name of the table that is already resident in memory. Or specifically, will be by the tame we call this DataRobot connection to have predications executed.

It's probably helpful at this point to back up a second and talk about the specific data being used for this example. It is a healthcare use case where we are calling DataRobot to make predictions about the likelihood a patient will be re-admitted to the hospital once they have been discharged. In this case, we have a table named DiabetesData that contains 50 different health variables. We used the Auto-Pilot feature in DataRobot to train many different models. Chose the "winner" and deployed it. Now we are pulling the most current information for a given set of patients and want to see what the readmission prediction is.

Regardless of whether we have populated our DiabetesData table with all patients currently in our hospital(s) and running this is a batch, or whether we are using an On-Demand Application Generation or Dynamic View setup, the DiabetesData table is the "Resident Table" we wish to pass to our DataRobot deployment so we need to enter that name. Then you can simply check the box for "DataRobot Predictions" (name given in the Configuration of the connection.) Lastly, press the "Insert Script" button.

Qlik Sense will then insert a simple ScriptEval call into your load script for you.

Now anytime the Load Script is executed, your DataRobot predictions will be available via the data model. Notice in the image below that our DiabetesData and DataRobot Predictions tables are associated based on that Master_Pastient_ID field that we identified in the configuration. We can now display our patient data and our predictions anyway that we wish in our application. With the knowledge that the predictions have been made based on the most recent information, we have in the application. If you are using an On-Demand Application Generation (ODAG) or Dynamic View strategy, it will be based on the live values read at the time and as frequently as the end user requests it.

Getting SaaSsy with DataRobot