DataRobot Predictions from Either Path You Choose

Dalton_Ruer · Sep 8, 2022 2:15:27 AM

In my recent Getting SaaSsy with DataRobot post, I documented how to use the DataRobot Analytics Connector from within your Qlik Sense applications. I know this will sound crazy but what if you want to make predictions on data you aren't loading into your application? Maybe you are collecting input parameters in your application from end users to play what-if games. Maybe you will record the predictions but you also want to take immediate action based on their values (ie prescriptive analytics.) Well, those things sound like perfect use cases for Qlik ApplicationAutomation.

It gets better my friends. Whether you are using a dedicated on-premise DataRobot server, a dedicated tenant or you are on the leading edge path with DataRobot's shiny new AI Cloud Manager using Paxata, Qlik Application Automation has you covered, and so do I.

In this post, I will help you identify the right DataRobot Connector Block to use for your path, help you understand how to execute predictions, and help you understand what to do with the output from the predictions.

Choose your Path

You have already chosen your DataRobot path, now it's just a matter of choosing the correct block from the DataRobot Connector. You probably would have guessed from the elaborate way I described the DataRobot choices which Qlik Application Automation block goes to which environment. But to be sure ... If you have a dedicated DataRobot server, or you have a dedicated tenant, you should use the List Prediction Explanation from Dedicated Prediction API block. If you are using the DataRobot AI Manager environment with Paxata, you should use the List Predictions block.

Oh no! What's that you are saying? You weren't told which path your organization chose, you were just given credentials and you just log in. Don't sweat it. I can help you with that. Just go to your Deployments within DataRobot, choose the Deployment you are going to execute predictions against, and choose Predictions, Prediction API and Real-time. DataRobot will provide all of the clues we need to choose the right block.

If the API URL contains App2.datarobot.com like in the first image below you are working with their AI Cloud and will need to use the List Predictions block. However, if you see a Dedicated path in your API_URL such as Qlik.orm.datarobot.com (second image) you will need to use the List Predictions fro Dedication Prediction API.

There are some other clues above as well. Notice in the second image the rest of the URL path contains api/v2/deployments, while the second image contains preApi/v1.0/deployments. It's basically DataRobot telling you which of their API's you need to utilize.

So how will that help you know for sure? One of the things that many people seldom look at with Qlik Application Automation blocks is the Description. Simply drag either/both of the blocks onto your canvas and scroll all the way down in the right panel. If you look at the List Predictions for Dedication Prediction API description you will see the following and notice it clearly indicates preApi/v1.0/deployments.

However, if you press the Show API endpoint link for the List Predictions block it will look like this. Both are dead giveaways as to the block/path you should choose.

Configuring your Connection

Regardless of which block you are using, you will first need to create a Connection for Qlik Application Automation to your DataRobot environment. If you have a Dedicated server your Connection details will look like this. Notice that you will need to copy the api_key value right from the DataRobot Deployment Details:

Your DataRobot AI Cloud connection will look similar and again you would need to copy your api_key from the deployment details. My DataRobot AI Cloud is just a "trial", hence I chose that region, while my dedicated tenant (above) is the US. The biggest difference is in the domain. Dedicated connections will be app.datarobot.com and AI Cloud connections will be app2.datarobot.com:

Choosing your Deployment

Once you test/save the connections we are ready to start making predictions. The List Predictions block is the easiest to set up so we will start with that one. Simply click the drop-down in the Deployment Id field, press "Do Lookup" and then choose the specific deployment model you are going to be making predictions against.

Then you simply provide the Input data you want to pass to the deployment to have predictions made for. More about the Prediction Data later, but for now notice that I've simply hard-coded a JSON string with field/value pairs:

The List Predictions fro Dedication Prediction API requires us to do a few more things that need to be completed. The first is the Dedicated Prediction Url. Good thing I had you bring up your deployment details because we will just copy it. Notice you do not need the HTTPS:// or the /predApi.. text, just the actual URL information.

Next will again simply click the drop-down for the Deployment ID, click Do Lookup and then choose our desired deployment.

Next, we copy the Data Robot Key from our deployment details and then we can insert our JSON block. Again, more later about that so don't panic in thinking I'm suggesting that you hand code the values you want to predict. It's just to make this section easier to navigate. 😁

Qlik Application Automation provides 4 additional parameters that are part of the DataRobot API specification. You can define the Passthrough Columns, Passthrough Columns Set, turn on Prediction Warnings and set the Decimals Number format.

You can refer directly to the DataRobot API documentation for all of the details you wish. For instance, notice that I have the Prediction Warning Enabled set to "true." Getting warnings sounded like a good idea. But alas, I ended up with an error.

Well, it turns out that in order to utilize the Prediction Warning Enabled there is work that must be done on our Deployment within DataRobot.

I guess I could have saved myself the trouble had I read the documentation. Oh well, I simply changed my default back to false so that the prediction can run.

Choosing your Prediction Input Data

Above I simply demonstrated the JSON format you need for your Prediction Data with hard-coded values. I've used my DataRobots and have predicted with a 99.99999999% confidence level that your goal in reading this isn't to hardcode 50+ input values each time you want a prediction. Instead, that data will come from somewhere else. Which is perfectly ok. Maybe you will be pulling the values from some other system as part of a workflow. When A event triggers this Qlik Application you will go do B and C and then assign output from those things to Variables that you will use as the Prediction Data. That's a great plan ... simply choose your variables, and use them where you need them in your Prediction Data. Notice I have already assigned the MasterPatientID variable and am in process of choosing the race variable below.

I'm so sorry. You don't like to use values, and you weren't doing A, B and C you were actually firing a SQL Query live based on input to your workflow and you wanted to use the data from the SQL Query. That is brilliant. Pulling the live data, when whatever event you have chosen triggers the Automation. You should write some posts. No problem, Qlik Application Automation will absolutely allow you to do that.

Or perhaps you are using a writeback solution, like Inphinity Forms, within a Qlik Sense application to capture input parameters and you wish to use those values. Do that.

Or perhaps you are ... You get the point. The Prediction Data simply needs to be a JSON block containing the field/value pairs. How you construct it, or read it from an S3 bucket, or pull it out of thin air doesn't matter. Which is the beauty of working with DataRobot within Qlik Application Automation.

Working with your Results

Woohoo, you now have a block that will execute a deployment in your DataRobot environment, regardless of which kind, and we are now ready for those wonderful predictions. Perhaps the first thing you noticed about the blocks List Predictions and List Predictions from Dedicate Prediction API was that they start with List as opposed to Get. It's of course because you may be passing a single row of data as Prediction Data or you could be passing many items in the JSON block. So these blocks are handled as lists, even if it is just a list of 1 prediction.

The DataRobot Connector for loading data into our applications simply returns the Prediction value, which is 0, in this case (the patient is not predicted to be readmitted.) However, notice below that within Qlik Application Automation either prediction block will return the Prediction as well, but it will also return a list of the Prediction values and the scores for each possible value. In my case, the 1, likely to be readmitted was scored at 0.0428973004, whereas the 0, the Prediction, was scored at .571026996.

Who cares?

Well, maybe you do. As I started this post I mentioned that perhaps we want to take action(s) based on the predictions which might be why you are making the prediction in your Qlik Application Automation workflow instead of just making the prediction in a Qlik Sense Application. If we are writing a flow that is "prescriptive" we might want to check the values. Ooooh .49999999999 vs .500000000001. Maybe that will be Action A, just email someone. While .000000001 vs .999999999 tells us that it's safe to go ahead and take the really expensive Action Z. So we might want to set up a Conditional expression.

Regardless of what we do with the values, Qlik Application Automation allows you to simply choose the values right from the block, just like it allowed you to choose Variables or data from another source.

What about Explanations?

If you don't already know me I will bring you up to speed quickly. I have very defined boundaries and am really particular about how things are worded. For example, take the phrase Data Science. Well, Science is explainable. Therefore, if something isn't explainable it isn't science. And if your predictions aren't explainable, then that isn't Data Science, it's just Data. One of the key reasons you are likely using DataRobot is the fact that it can so wonderfully return explanations for its predictions.

The Prediction of 0 above is nice. But knowing what factors led to the prediction may be just as valuable when helping us choose our prescriptive actions. Well, my friends, Qlik Application Automation has you covered for that scenario as well. In fact, you can see from the following image the block is literally raising its hand and begging you to choose it. List Prediction Explanation from Dedicated Prediction API will give you not only the Prediction, and the Prediction values but it will also return the explanations to you.

Wait something must be wrong. I see a qualitativeStrength of +++, but the second is --. What do those mean? Oh yeah, now I remember ... Qlik Application Automation is just calling the provided DataRobot API's so I might as well check the documentation from DataRobot so that I get a full and complete understanding of the input paramters I can choose for the block and understand the output values. Sure enough, it's covered.

https://docs.datarobot.com/en/docs/api/reference/predapi/pred-ref/dep-predex.html

Bonus Block

I see you out there on the leading edge doing Time Series Predictions in DataRobot. Not an issue, Qlik Application Automation has you covered with a block as well. Simply choose the List Time Series Predictions from Dedicated Prediction API and you will be good to go.

The initial inputs needed are already covered above. However, there are a few additional parameters you will need to input as well.

Of course, DataRobot has you covered with complete documentation at https://docs.datarobot.com/en/docs/api/reference/predapi/pred-ref/time-pred.html

The information in this article is provided as-is and to be used at own discretion. Depending on tool(s) used, customization(s), and/or other factors ongoing support on the solution below may not be provided by Qlik Support.

DataRobot Predictions from Either Path You Choose