Qlik AutoML: How to use Coordinate SHAP Table

[KellyHobson]

In a previous article (Qlik AutoML: Overview of SHAP values), I described Shapley values and what they represent for AutoML models. These additional measures at the record level can be helpful to enhance data visualizations and understand a model further.

A helpful trick for using Shapley values in Qlik Sense sheets is to generate the Coordinate Shapley table. This table turns the feature to a record level measure. Thus providing additional methods for aggregating and exploring the values.

In the steps below, I'll outline how to generate this table. I'll be using the HR Analytics: Promotion Data. This dataset has series of features used to predict whether or not an employee will be promoted.

Steps

1. Upload train.csv and test.csv from HR Analytics data to Qlik Catalog.
2. Create a new ML Experiment and select train.csv as the training dataset. Set "is_promoted" to the target variable and unselect "employee_id" as a feature. Run the ML experiment.
   
   
   
   
3. Click on 'Deploy' and then follow the link at the top of the webpage which will take you to your deployed model within the "ML Model Management" interface.
4. Within "Deployment Overview", select "Create Prediction".
   
   
   
   
5. For the apply dataset, select test.csv from Qlik Catalog. 
   
   
   
   
6. Under Prediction Options, include "Coordinate SHAP" table and set "employee_id" as your index column. Then click 'Save and Predict Now.'
   
   
   
   
7. Within Catalog, you will now see the prediction tables and Coordinate SHAP table available.
   
   
   
   If you open up the Coordinate_SHAP table in Catalog, you can see the different structure than the SHAP table.
   
   
   
   
   
   
8. Open a new Sense App and upload the prediction tables generated in step #7.
   
   
   
   I applied all recommended associations, so the tables are linked by "employee_id."
   
   
   
   
9. Add a new horizontal bar chart to a sheet with 'automl_feature' as the dimension and the average 'SHAP_value' as the measure.  This way you can review whether the feature influence the predicted value in a positive or negative way. Additionally, you can add filter dimensions to compare different groups within the dataset. In the example below, we compare 'R&D' vs 'Finance' Department. 
   
   
   
   

Environment

Qlik AutoML 

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.