Skip to content

Latest commit

 

History

History
 
 

cs

README.md

Product Recommendations Sample

The sample under the Recommendations.Sample folder is a C# sample that shows how to use an API endpoint provisioned by the Recommendations preconfigured solution. You can find more details about the Recommendations preconfigured solution here: https://aka.ms/recotemplate. The sample uses an API client autogenerated using AutoRest using the Recommendations service swagger file availble under https://<service_name>.azurewebsites.net/swagger/docs/v1 .

The sample demonstrates how to:

  1. Train a product recommendations predictive models based on historical transaction data and information on the product catalog
  2. Get Item-to-Item Recommendations from a trained model
  3. Get Personalized Recommendations from a trained model
  4. Set a model to be the default model, and request recommendations from the default model

Pre-Requirements

The steps below are explained in more detail in the Getting Started Guide

  1. Before you can run the application you need to deploy the Recommendations Template. See Deployment Instructions for more detail

  2. You will get an API Key as part of the deployment process and the url of your solution. Set the apiAdminKey and recommendationsEndPointUri variables in Recommendations.Sample/Program.cs file based on those values

  3. You will need to upload your catalog and usage events data for training to some container in the service's blob storage account

Note that training a recommendation model is performed using a single catalog file (optional) and at least one (but potentially multiple) usage events files.

Any usage events files need to be uploaded to a folder under the blob container. The path of that folder will have to be specified as the 'usageRelativePath' parameter in the train new model API request. It is also possible to set 'usageRelativePath' to a single specific usage file.

Similarly, any usage events file(s) used for evaluation, should be uploaded to a separate folder under the blob container. The path of that folder will have to be specified as the 'evaluationUsageRelativePath' parameter in the train new model API request. It is also possible to set 'evaluationUsageRelativePath' to a single specific usage file.

The catalog file can be uploaded directly under the container or nested under some subfolder(s). The path to the catalog file (including its name) will have to be specified as the 'catalogFileRelativePath' parameter in the train new model API.

As a summary, the following parameters are used to specify the locations of the various data inputs to model training process:

  • blobContainerName
  • trainUsageRelativePath
  • trainCatalogFileRelativePath (*optional)
  • evaluationUsageRelativePath (*optional)

Learn more about model training parameters here..