The following provides deeper information on the Kraken | Qlik integration, along with a series of best practices to maximize the workflow between Big Squid’s Kraken platform and Qlik
Key Features of the Kraken | Qlik Sense Integration
Kraken requires a url, username, password to connect to your Qlik Sense environment. Authentication methods currently supported by Kraken include default Windows authentication and Active Directory.
Kraken users can add multiple Qlik Sense environments by adding them as separate Data Providers in Kraken. Each Qlik Sense connection can be given a unique name so they can easily be identified throughout Kraken.
Once connected to Qlik Sense from inside Kraken, Kraken can:
List all apps you have available in your personal workspace (“Select dataset” screen)
For apps with multiple, unassociated data tables, Kraken will list each table as separate datasets (ie islands)
Upon selecting your dataset, it will retrieve the underlying data model from any app you select and return the schema. Depending on the size of the app, the data retrieval can take a few minutes.
By selecting a metric and “analyzing”, Kraken will analyze the data model to produce multiple machine learning models from it
Deploying a model and creating a Predicted Dataset in Kraken will produce predictions on the input data that was used to create the Analysis. Kraken creates a new dataset in .csv format will provide a download URL that can be added to any Qlik app via creation of a new Web File data connection in the Load Script. Important things to note:
Kraken uses the full starting input data from the Qlik app and appends predicted columns.
Kraken provides a URL to the user that can be used to create a new Web File data connector in Qlik. For more information on how to create a Web File connector and add it to your Load Script, refer to the Qlik documentation: Qlik Web File Data Connections
If the “keep up to date” feature is toggled on inside Kraken, a fresh .csv will be generated and replace the existing Predicted Dataset once per day.. If data in the app’s underlying data model changes or new records added, they will be added to the output file with predictions appended to it
Best Practices in Using Qlik Sense & Kraken
If using a published app to build models in Kraken, we recommend copying the app to your personal workspace, as this is the source Kraken uses to list available Qlik Sense apps.
The Qlik Sense Associative Engine is a powerful and beneficial means to join related data together. It is a key aspect of the Kraken integration, as the same curated, associated data is available to gain visibility into key metrics in the app. The downside is the associations may contain data unnecessary to model the intended use case in Kraken. We recommend:
Removing any unneeded tables or data in your Qlik Sense app
Structuring the app’s data model at the granularity you desire for the predicted outcome
We highly discourage the use of synthetic keys in a data model to optimize the performance of retrieving datasets from Qlik Sense
We recommend using Kraken’s “Refine” feature both before and after analysis to reduce the number of model inputs to those that are relevant for the use case. This follows a best practice in machine learning in the law of parsimony.
When loading the .csv file with Kraken predictions into your data model, we recommend only loading in the columns you need for the data model. Often this is the unique identifier, along with the predicted columns. This will avoid duplication of fields in the data model.
Key Functions and Features of using Kraken with Qlikview
Kraken supports Qlikview by way of a “file upload” data source
Once a model is built in Kraken, the predictions can be downloaded to a local directory
Best Practices in using Kraken with Qlikview
With Qlikview, best practice is to create a table view of the data you wish to bring into Kraken
Then export that view of the data into a .csv format
Load this .csv file into Kraken
Download predictions to a local file directory for load into Qlikview
Users can possibly extend the utility of this workflow through Qlikview’s scheduling features