Predictions

You can create manual or scheduled predictions for various problem types in the Rapid Model recipe, including binary classification, clustering, time series forecasting, anomaly detection, regression, and multi-class classification.

When you create a prediction job, the platform automatically performs data preprocessing and feature engineering on the uploaded dataset. This step ensures the data is refined before being fed into the trained model, improving the accuracy of predictions.

Types of Predictions

Manual Prediction – Select a model and upload a new dataset each time you want to generate predictions.
Scheduled Prediction – Set up a scheduler to run predictions at a defined frequency without manual intervention. You can configure it to use an updated dataset, ensuring predictions are always based on the latest data.

Creating a manual prediction

Use this procedure to create a prediction run in a project.

Hover over the menu icon and select Projects. The Projects dashboard is displayed.
Select the project for which you can to create a manual prediction. You can only create manual prediction for the data pipelines which have models built.

Click the Predictions from the project-level left navigation menu to open the predictions page. If the prediction is created already, you can view the list of prediction runs, else click +Manual Prediction.

Note

This + Manual Prediction button is enabled only when there are models created in the project.

This opens the Select model to generate predictions window.

Select the model that you want to use to generate predictions on the new dataset.
Click Proceed to go to the next step.
By default, the model name is populated and cannot be edit. On the breadcrumb, you can provide a custom name for the prediction run name, else the system generated name will be considered.
Click Browse to upload the new dataset on which you want to perform predictions. This opens the File Upload window.

Click Upload File From Local. You can upload maximum file size of 50MB.
Click Close once uploaded.
Click Generate Prediction to start the prediction process. You can notice the status of the prediction job is set to Created and this later changes to Running once the prediction job starts.

Click Add To Canvas to add the generated dataset with predictions to the canvas. This opens the Update Name dialog where you can provide the prediction output name and click Add to Canvas.

Click Download to download the prediction file to your local system to view the predictions generated by the model.

Note

The Future Period field is displayed when creating a prediction run for time series problems. In this field, you specify the future period for which predictions should be made.

On this page, you can perform the following actions:

Click the logs icon to check for the errors if the prediction run failed to execute. You can also view logs for the successfully executed prediction jobs.
Click the Caret icon and select Delete to delete this prediction run.

Managing manual predictions

Use this procedure to manage all the manual prediction runs in a project.

Hover over the menu icon and select Projects. The Projects dashboard is displayed.
Select the project for which you can to create a prediction run. You can only create prediction runs for the data pipelines which have models built.
Click the Predictions from the project-level left navigation menu to open the prediction jobs page. You can see the list of prediction jobs created.

Review this information:

Name:

The name of the manual prediction job.

Triggered By:

The user who has executed the prediction job.

Updated at:

The last updated time stamp of the prediction job.

Input model:

The name of the trained input model used on the new dataset to make predictions.

Status:

The status of the prediction job after it is run. Possible values:

Success
Failed

Predictions:

Click the download icon to download the dataset in which you can view the predictions.

Log:

Click View to view the logs generated for the job.

Click on Table Settings to customize your table view by selecting or deselecting the columns you wish to display. You can also rearrange the columns to set them in your desired order.

Creating a prediction scheduler

Use this procedure to create a prediction scheduler for a model within a project.

Hover over the menu and select Projects. The Projects dashboard is displayed.
Select the project for which you want to create a prediction scheduler. You can create prediction schedulers for different models in a project.

The Canvas page is displayed.

Click the Predictions tab on the project navigation menu on the left to open the Predictions page.
Do one of the following:

Click the plus icon |ico1| on the top right-corner of the page.

Click the +Prediction Scheduler option to create a predictionscheduler. However, you can only view this option when there are no schedulers created in this project.

The following page is displayed where you can create a prediction scheduler to run this data pipeline at the set time interval.

Select the model that you want to use to generate predictions at the scheduled frequency.
Click Proceed.
Select the model on which you want to generate predictions at the scheduled frequency.

Select the scheduler frequency. Possible values:

Daily - This displays Hrs and Min. drop-down to select the time at which the job should be triggered.

Weekly - This displays days in a week and time at which the scheduler should be run.

Cron - This displays the unix-corn format to create a scheduler.

View the data pipeline on the canvas.
Click Save to create the prediction scheduler. This also enables the +Destination option to configure the data connector to which you can publish the generated output datasets or the input dataset.

You can see the project variables button only if the variables are defined at the project level. After creating the scheduler, you can change the value in project variables.

Click + Destination. This opens the Destinations side panel.
Click + Destination.
Select the dataset that you want to add to the destination. If the dataset list is huge, you can use the search option to search for the dataset you want.
Select the destination from the drop-down list. You can only view the list of external data sources configured under this tenant excluding Snowflake and Fivetran connectors.

When you select the SQL connector to synchronize or copy the output dataset generated after running the project, the table name column is displayed. Here, you can provide the table name and select either “Append” or “Replace”. Opting for the “Append” option will append the dataset to the existing one, provided both datasets share the same schema. Alternatively, selecting the “Replace” option will replace the existing dataset with the new one.

If you choose the data connector as MongoDB, you can provide the database name and collection. In the event that the provided collection name already exists, the new dataset will be appended to the existing collection.

Provide the destination folder and destination file name to save the file in the destination folder with the new file name after the job is run every time at the scheduled time.
Click Save to save this destination. This button is enabled only after you select all the required destination fields.

Note

You can store files in multiple destinations. To add another destination, click + DESTINATION. If you want to remove any destination, click the delete icon.
If you no longer want to save the output to the configured destination, you can use the delete icon to delete the destination.

Close the window after configuring the destination for the job.
Click GLOBAL VARIABLES to change the configured parameters for this job.

Note: The GLOBAL VARIABLES button is enabled only when the global variables are declared at the project level. To configure global variables, refer configuring global variables at a project level.

Change the value for the key. Please note that you cannot change the key.

Running the scheduler manually

Use this procedure to manually run or re-run the prediction scheduler. However, scheduled runs occur automatically based on the configured recurrence.

To manually run a prediction scheduler:

Select the project in which you want to run the job.
Click the Predictions tab to view the list of manual predictions and prediction schedulers for this project. The predictions list page is only visible if predictions have been created for the selected project.
Click the scheduler name link you want to run manually. This opens the specific prediction scheduler’s page.
Click Run to initiate the job manually. This opens the Manual Run Configuration side panel.
Enter a run name and click Run. Once the prediction scheduler run starts, its status changes from Created to Entity Loading and then to Running. When the run completes successfully, you can view the output on the Run History page.

To re-run a job, click the ellipsis icon in the Run Name column on the Run History page and select RE-RUN.

To delete a specific run, select DELETE.

Managing Prediction schedulers in a project

Use this procedure to manage all the prediction schedulers in a project.

Hover over the menu icon and select Projects. The Projects dashboard is displayed.
Select the project for which you can to schedule or create a job. You can create jobs for different scenarios in a project.
Click the Predictions tab on the left navigation menu of the project to open the schedulers page and view the list of manual and automatic schedulers you have already created.

Note

If there are multiple schedulers, you can use the search option to find the scheduler you want.

You can also create a new scheduler, using the plus option.

Click on the Prediction Scheduler name that you want to edit. This redirects you to the Prediction scheduler page where you can edit the prediction scheduler details.
Modify the required details.
Click Save to view the new changes.

On this Prediction scheduler page, you can also:

Run this job manually, clicking the Run button.
View the run history, using the Run history icon. This allows you to view the history of all prediction scheduler runs till date and up to 300 records of last 30 days.
Pause the job that is running, using the Pause icon. You can click the same icon to start the paused prediction scheduler.
Click the Action drop-down to select the Delete option to delete this prediction scheduler permanently.
Click the Timeout 1hr option to change the time out duration of the prediction scheduler. You can view this option when you click the Actions drop-down. By default, the time out duration is set to 1 hr. Setting this will terminate the prediction scheduler after this duration.

Publishing the updated data pipeline to selected prediction scheduler from canvas

Use this procedure to republish the data pipeline to prediction scheduler. When you update the dataset, delete a recipe or add a new recipe to the data pipeline, you can republish the new flow to the prediction scheduler using the Publish to Prediction Schedulers option on the canvas. This updates the canvas on the selected schedulers.

To publish the changes made in the data pipeline to all or specific prediction scheduler(s) in a project:

Select the project to navigate to the canvas view page.
Click the Actions drop-down and select Publish to Prediction Schedulers on the canvas. This displays Republish Model Flow to Prediction Scheduler dialog.

This displays the list of prediction schedulers to which you want to publish the latest or updated data pipeline.

Select the check boxes corresponding to the prediction schedulers to which you want to update the latest canvas. This enables the Yes, Republish button.
Click Yes, Republish to republish or update the latest data pipeline to the selected prediction schedulers.

From the next schedule, the prediction scheduler run is performed on the new modeling pipeline.

Fetching the latest data pipeline to a specific prediction scheduler

Use this procedure to fetch the changes made to the data pipeline on the canvas to the data pipeline in a specific prediction scheduler.

To publish the changes made to the data pipeline on the canvas to a specific scheduler from the prediction scheduler page:

Select the project to navigate to the canvas view page.
Select Predictions from the project level navigation. This takes you to the Schedulers page where you can view the list of schedulers created for this project.
Select the prediction scheduler to which you want to publish the changes made to the data pipeline. This takes you to the selected prediction scheduler page.
Click the Republish button in the canvas section to incorporate all the changes that were made to the canvas at the project level to this pipeline.

The Republish Model Flow to Prediction Scheduler window appears.

Click Yes, Republish to republish the project canvas to the scheduler.

Comparing the canvas of the prediction scheduler with current canvas of the project

Use this procedure to compare current canvas of the project and canvas of the prediction scheduler side-by-side to track changes.

To compare the canvas of the scheduler with the current canvas of the project:

1.Select the project to navigate to the canvas view page.

Select Predictions from the project level navigation. This takes you to the prediction schedulers page where you can view the list of schedulers created for this project.
Select the scheduler that you want to compare with the current canvas of the project. This opens the scheduler page.
Click Compare to compare the canvas of this prediction scheduler with the canvas of the project to notice the differences. You can see the difference between both the canvases side by side.

If you notice the canvas of the prediction scheduler is not up-to-date, you can click Republish to fetch the latest canvas of the project to update the changes.