A build of a model is created when data files have successfully been loaded to the model container. This procedure is called “creating a model build” or just “creating a build”, and it starts the training process of the machine learning an algorithm that generates a build.
After a build is created the model is ready for recommendation.
- To see and work with builds of a given model click the
icon, enter Models, and select the relevant link. Then click the Builds action.
A recommendation model must contain at least one active build to make product recommendations. At some point in time, the recommendation model in use should be updated with more recent transaction data. This is done by creating a second build including the new usage data acquired from the date when the first build was created. This process should be repeated on a regular basis as new data is acquired.
See Scheduler Jobs for more information on how this procedure can be automated. This topic, however, focuses on how to create a build manually.
Note: If a model has been successfully created using the Create & Build Model with Data action on the Model card page or by following the three steps in Actions - Model Procedure, the first model build should be ready.
Build parameters
The build parameters values cannot be adjusted for an existing build and can only be edited when a new build is created or in the Model Template.
The General FastTab has the following fields:
| Field | Description |
|---|---|
| Build Description | Displays the description of the build. If the build is not created manually it will get the description that is combined with the text Default Build No. plus the model's build number. |
| Is Active Build | The active build is the build that is used by the model to give recommendation based on the model data. This check box indicates if the build is the model's active build. |
| Build Start Time | The date and time when the build procedure was started. |
| Build Status | Displays the current status of the build. The possible status values are: New: The build only exists in LS Central. The build procedure has not been initiated in Azure. Created In Azure: The building of the model in Azure has not yet started. In Progress: The model is being built in Azure based on the current data. Completed: The building of the model was successful and the build is ready to make recommendations. The action Set as Active Build is used to assign the build to be the one that the model uses to give recommendation. Failed: The building of the model has failed in Azure. |
| Status Message | More detailed information on the Build Status. |
The values in the last three fields are retrieved from Azure by the actions Get All Builds and Get Build in Departments - LS Retail - LS Recommend - Models - Builds, under the Azure Data action group).
The remaining fields in the Build Parameters FastTab get their values from the Model Template, which contains parameter default settings for a recommendation model. These parameters are described in subsections of the Model Template and can be adjusted without overwriting the default settings when a new build is created.
Create a new build
- Click the
icon, enter - Select a model that has uploaded a catalog and usage data, that is a model with Model Status Ready to Build or Ready for Recommendation.
- Click the Builds action to open the list page, and then click New to create a new build.
- Select the build parameters as explained above.
- Click Actions - Build action to trigger the building process.
- Click Actions - Get Build to refresh the information on the Build Status. This will retrieve the build data from Azure. Note: It is good practice is to refresh the build status before interacting with the build.
- When the Build Status becomes Succeeded, the build process has finished and you can make the build the model's active build by clicking Model - Set as Active Build.
Actions
The following actions are available on the Actions menu on the Build card page:
- Delete: This action deletes the current build in Azure and from the model container.
- Build: Clicking this action triggers the building process in Azure based on the selected parameters and the uploaded data.
- Build - Get Build: Click this action to refresh the build data from Azure. A good practice is to refresh the build data before interacting with the build. This action populates the fields Build Start Time, Build Status, and Status Message. In addition, the Features list is populated if features have been included in the model's catalog.
- Model - Set as Active Build: Clicking this action makes the current build active in the selected model. It means that LS Recommend will use the current build of the model to make recommendations.
- Model - Features: Clicking this action will show all features used in the current build. See Features for more details.
- Errors: Displays errors, if any, that occurred during the building process.
Recommendation options
LS Recommend supports two recommendation options that can enhance product recommendations to the consumer. These options have attractive capabilities that are activated by feeding more information than necessary to create a model.
- Get Personalized Recommendations
This type of recommendation is based on a specific customer's or member's usage history, which means that items can be recommended to users despite an empty basket. These recommendations can help users to discover items that they may be interested in. This is only useful if a history exists for customers or members in the database. The Personalized Recommendation check box in the Model Template or the Build card page must be selected in order to train the model with user specific information.
-
Use Cold Item Placement
LS Recommend can use features to improve product recommendations for new items, or items with little sales history, so-called cold items. In order to use this option the Include Features In Catalog check box in the Model Template or the Model card page must be selected. This means that additional information is provided to the model about the items in the catalog, called features.It is also possible to allow for recommendations based on the similarity between pairs of cold items by selecting the Cold To Cold Item Recommendation check box in the Model Template or the Build card page. Cold To Cold Item Recommendation can only be selected if Include Features In Catalog is selected for the model.