Implement custom prediction model

Level: advanced

note

The functionality is relevant to Classic UI.

To implement the example:

1. Add an ML model. Read more >>>
2. Train the model. Read more >>>
3. Implement prediction business logic. Read more >>>

Example

Implement automatic prediction for the Category account field based on the values of the Country, No. of employees and Industry fields while saving the account record. Implement the following prediction conditions:

• Create a model that learns from account records for last 90 days.
• Re-train the model every 30 days.
• Set permissible prediction accuracy for the model to "0.6."

To implement the example, you can also use page fields and filters, as well as the Predict data system action element in the business process. Learn more: Predictive data analysis (user documentation).

1. Add an ML model

1. Open the Configuration section. Instructions: Open the Configuration section.

2. Create a sdkCustomPredictionModel package. Instructions: Create a user-made package using Configuration section.

3. Change the current package to sdkCustomPredictionModel. Instructions: Change the current package.

4. Click in the top right → System setup → Lookups.

5. Open the ML model (MLModel code) lookup.

6. Click New → Lookup prediction to add the lookup value.

7. Fill out the properties of the lookup value.

   Property	Property value
   Name	Predict account category
   Type	Lookup prediction
   Object	Category of Account

8. Click Next. This opens the Predict account category page.

9. Fill out the parameters of the prediction model.

   Property	Property value
   Prediction enabled	Select the checkbox
   Parameters → Automatic model training settings
   Quality metric lower limit	0.60
   Retrain after, days	30
   Advanced settings → Advanced tools to add columns
   Query for selecting additional training data	new Select(userConnection) .Column("a", "Id") .Column("a", "CountryId") .Column("a", "EmployeesNumberId") .Column("a", "IndustryId") .Column("a", "AccountCategoryId") .Column("c", "Name").As("AccountCategory") .From("Account").As("a") .InnerJoin("AccountCategory").As("c") .On("a", "AccountCategoryId").IsEqual("c", "Id") .Where("a", "CreatedOn") .IsGreater(Func.DateAddDay(-90, Func.CurrentDateTime()))
   Query metadata for selecting additional training data	{ "inputs": [ { "name": "CountryId", "type": "Lookup", "isRequired": true }, { "name": "EmployeesNumberId", "type": "Lookup", "isRequired": true }, { "name": "IndustryId", "type": "Lookup", "isRequired": true } ], "output": { "name": "AccountCategoryId", "type": "Lookup", "displayName": "AccountCategory" } }

10. Click Save.

11. Click Close. This opens the page of the ML model (MLModel code) lookup.

2. Train the model

1. Select the Predict account category.

2. Click Actions → Execute model training job.

   Wait until the Status field value is set to "Done." The process might take several hours. It depends on the amount of passed data and general workload of the predictive service.

3. Implement prediction business logic

1. Create a business process

1. Open the Configuration section. Instructions: Open the Configuration section.

2. Select the sdkCustomPredictionModel package.

3. Click Add → Business process.

4. Open the Settings tab.

5. Fill out the schema properties.

   Property	Property value
   Title	Predict account category
   Code	UsrPredictAccountCategory

6. Click Save.

As a result:

• The "Predict account category" business process will be created.
• Creatio will add the "Predict account category" business process to the Process library section.

2. Set up the business process parameters

1. Click an arbitrary place in the working area of the Process Designer.

2. Open the Parameters tab.

3. Add the business process parameter.

   1. Click Add parameter → Lookup.

   2. Fill out the parameter properties.

      Element	Element type	Property	Property value
      Parameter that contains the model ID	Lookup	Title	ML model ID
      		Code	MLModelID
      		Data type	Lookup
      		Lookup	ML model
      		Value	Click → Lookup value → Predict account category → Select. This populates the property using the "[#Lookup.ML model.Predict account category.99402097-a6ac-4930-9540-6a5cb8d6d9b2#]" value.
      Parameter that contains the account ID	Lookup	Title	Account ID
      		Code	AccountID
      		Data type	Lookup
      		Lookup	Account

4. Click Save.

3. Set up running the business process

1. Delete the Simple page element.

2. Click .

3. Place the Throw signal element in the working area.

4. Click → connect the elements.

5. Click → Wait for signal.

6. Fill out the element properties.

   Property	Property value
   Title	Run the process

7. Fill out the element properties.

   Property	Property value
   Which type of signal is received?	Record signal
   Record Id	Click → Process parameter → Process parameters tab → Account ID → Select.
   Which event should trigger the signal?	Record added
   The added record must meet filter conditions	Click Add condition. This opens the Select column window. Select "Country" in the Column property. Click Select. Click → "is filled in." Repeat steps 1–4 for both No. of employees and Industry columns.

8. Click Save.

4. Implement populating the Category account field

1. Add a Script task element. To do this, click → place the Script task element between the Throw signal and Terminate page elements.

2. Fill out the element properties.

   Property	Property value
   Title	Call the automatic prediction

3. Implement the logic of working with process parameters. To do this, go to the element setup area and add the source code.

   Call the automatic prediction script task

   // IMPORTANT: When implementing
   // long-running operations,
   // it is crucial to enable timely and
   // responsive cancellation. To achieve
   // this, ensure that your code
   // is designed to respond appropriately
   // to cancellation requests using
   // the context.CancellationToken
   // mechanism. For more detailed
   // information and examples,
   // please, refer to our documentation.
   
   /* Get an instance of the "UserConnection." */
   var userConnection = Get<UserConnection>("UserConnection");
   /* ID of selected account. */
   Guid entityId = Get<Guid>("RecordId");
   /* ID of selected ML model. */
   var modelId = Get<Guid>("MLModelID");
   var connectionArg = new ConstructorArgument("userConnection", userConnection);
   /* Create an instance of the "MLEntityPredictor." */
   var predictor = ClassFactory.Get<MLEntityPredictor>(connectionArg);
   /* Call the prediction service. Creatio saves data to the "MLPrediction"
   database table. If high probability is predicted, Creatio saves data
   to the required field on the account page. */
   predictor.PredictEntityValueAndSaveResult(modelId, entityId);
   
   return true;

4. Click Save.

As a result, the prediction will be performed for new accounts and displayed on the account edit page.

This implementation of the prediction slows down saving an account record because because calling the prediction service is executed in 2 seconds. This can reduce the performance of bulk operations with data saving, like Excel import.

Source code

Call the automatic prediction script task

// IMPORTANT: When implementing
// long-running operations,
// it is crucial to enable timely and
// responsive cancellation. To achieve
// this, ensure that your code
// is designed to respond appropriately
// to cancellation requests using
// the context.CancellationToken
// mechanism. For more detailed
// information and examples,
// please, refer to our documentation.

/* Get an instance of the "UserConnection." */
var userConnection = Get<UserConnection>("UserConnection");
/* ID of selected account. */
Guid entityId = Get<Guid>("RecordId");
/* ID of selected ML model. */
var modelId = Get<Guid>("MLModelID");
var connectionArg = new ConstructorArgument("userConnection", userConnection);
/* Create an instance of the "MLEntityPredictor." */
var predictor = ClassFactory.Get<MLEntityPredictor>(connectionArg);
/* Call the prediction service. Creatio saves data to the "MLPrediction"
database table. If high probability is predicted, Creatio saves data
to the required field on the account page. */
predictor.PredictEntityValueAndSaveResult(modelId, entityId);

return true;