ML Model Definitions (Machine Learning)

You can configure Machine Learning by defining the models to be built. Building a model is also known as “training" a model. Models are trained with learning data and a target variable.

• The target variable is the field that you wish to gain insight into, predict, perform what-if on, detect anomalies on, or forecast.  
• The learning data is the data that you suspect has impact on the target field, the interval that you wish to forecast, the fields to input into what-if scenarios.

Define a new Machine Learning model

To begin defining a Machine Learning model, do the following: 

1. In version 11.1 and earlier, go to Main Menu > Admin Settings.
   In version 11.2, go to the left toolbar and select Administrator Settings > ML Models.
2. Select the ML Models tab. Before any models are created, the tab looks like the one below:
   
3. Click Create to begin defining a new model.

Dataset

The first step is to identify the data that you want to analyze using Machine Learning.

Select the name of the Dataset you want to analyze. You can also enter a where clause in the Criteria field to constrain the volume of data that will be analyzed.

Click Next.

Target

In the Target field, enter the specific field you wish to gain insight into. Note that the example below is not a time series example. For more information on time series, which involves forecasting one or more values out into the future, see the separate Time Series Example example at the end of this article.

Click Next.

Variables

Next, help identify the fields that may impact your Target field. What are the potential driving factors behind your target? Select these factors from the Available Properties list.

Click Next.

Name

Enter the Model Name and a Description.

Click Next.

Model retraining

After the initial build, models must periodically be retrained. Select the training schedule frequency of the model you just created. You can choose days or hours. Enter the number of days or hours in the space provided.

Click Finish to save the new model. Model training begins automatically.

After the model is defined

Once the models are built, you can run Machine Learning jKQL queries that perform analytics such as:

• FeatureSelection to determine driving factors
• Forecast for forecasting into the future and detecting anomalies
• Expected for finding out what a target is most likely to be given the driving factors
• Running what-if scenarios

Time Series Example

When dealing with time series data, you will need to build time series models. They're different from regular models in that they involve time. To make a model time series, in the Time Series field, indicate whether you wish to work with data in a time series manner (Yes or No).  You should specify data as time series if you wish to forecast into the future at a specified interval. In the example below, the pageViewDate is forecast once a day.

For time series models only, an Anomalies step is presented as part of the wizard. You must configure anomalies detection using the fields provided.

• If you want anomalies to be processed in real time, when they occur, select the Use Real Time Processing check box.
• Select the Anomaly Margin Type: Percent or Numeric.
• If you chose Numeric, enter the margin of error for a numeric value to be considered an anomaly in the Anomaly Margin field.

Generate a dashboard from a Machine Learning Model

You can generate a Machine Learning dashboard based on a jKQL Script like the one below.

1. In the jKQL Script Parameters area on the lower right, be sure to specify the dashboard, repository, organization, and ML model information, as shown in the green box below.
   

The RepositoryID must be the name of the repository followed by a dollar sign ($) and the Organization name. In the example above, BloodPressure$Nastel refers to the BloodPressure repository for the Nastel organization.

2. Click Test to make sure the ScriptResult is successful.
3. Click Save to save the script. Close the Script editor.
4. Create a temporary viewlet. At the jKQL prompt, enter get MLModel to return the Machine Learning model from which you want to create a dashboard.
5. Select the check box for the model and select Generate Dashboard from the pop-up menu.
   
6. A new dashboard is generated based on the model.
   
7. When the dashboard is complete, a confirmation message is displayed.
   
8. Click OK.
9. Save the dashboard.