Batch Predictions
Batch Prediction allows you to efficiently generate predictions for an entire dataset using one of your trained models. Instead of inputting data row by row, you select an existing dataset, and the system processes all records in the background, making it ideal for large-scale scoring tasks.
Use Batch Prediction when you need to score a customer base, evaluate the potential outcome for a list of leads, apply a model to historical data for analysis, or any scenario requiring predictions on multiple records simultaneously. The results are stored and can be reviewed later.
Batch Prediction Process
Automatic summary generation for prediction batches
Accessing Batch Predictions
Similar to Real-Time Predictions, you access the Batch Prediction feature from a specific model's detail page:
- Navigate to the Models section.
- Select the trained model you want to use for batch processing (e.g., `Model ID123`).
- Find the Predict or similar option, leading to the Prediction Interface page.
- On the Prediction Interface page, click the button labeled Batch Prediction.
This will reveal the components needed for selecting a dataset and managing batch prediction jobs.
Interface Anatomy
When Batch Prediction mode is active, the interface typically includes these components:
Dataset Selection & Initiation
This area allows you to choose the input data and start the process:
- Dataset Selection: Uses a component (like `DatasetUploadComponent`) to browse and select an existing, compatible dataset from your project.
- Selected Dataset Display: Shows the name of the chosen dataset (`dataset.name`). May include an icon indicating if the dataset is preprocessed (
dataset.preprocessed). - Predict Button: Clicking this initiates the batch prediction job for the selected dataset using the current model. (Ref: `BatchPrediction.vue`)
Ongoing & Completed Batch Predictions
This section (Ref: `HistoricalBatchPredictions.vue`) tracks the status of your batch jobs:
- Ongoing Predictions List: Displays jobs currently being processed.
- Completed Predictions List: Shows jobs that have finished (either successfully or with errors).
- Batch Card: Each list item shows key info: submission date (`batch_date`), average prediction output (`average_output`, if applicable), and current status (`status`).
- Status Indicator: A visual cue (e.g., colored circle , , ) next to the status text (`ongoing`, `completed`, `failed`).
- Real-time Updates (): The lists update automatically as jobs progress, receiving information via WebSockets (listening for `batch_prediction_update` events).
- View Details: Clicking on a batch card opens a modal with detailed results.
Batch Details Modal
Opened by clicking a card in the historical lists, this modal (Ref: `BatchPredictionDetails.vue`) provides comprehensive information about a specific batch run.
It includes summary statistics, input data overview, and paginated individual prediction results. (See Viewing Batch Results section below for more details).
Starting a Batch Prediction
Initiating a batch prediction job involves these steps:
Ensure "Batch Prediction" Mode is Active
Verify you are in the batch prediction view. If necessary, click the Batch Prediction button.
Select Input Dataset
Use the dataset selection component (`DatasetUploadComponent` provided by the parent `PredictionInterface.vue`) to choose the dataset containing the records you want to predict on. The selected dataset's name will appear.
Click "Predict"
Click the Predict button within the "Batch Prediction" section (Ref: `BatchPrediction.vue`). This sends a request to the backend (e.g., POST /batch_predict/{dataset.id}) with the model_id to start processing the selected dataset.
Monitor Progress
The batch job runs in the background. You can monitor its progress in the Ongoing Batch Predictions list. Once finished, it will move to the Completed Batch Predictions list.
Monitoring Batch Runs
The `HistoricalBatchPredictions` component provides visibility into your running and completed jobs:
Ongoing List
Shows jobs currently in progress. Their status will typically be "ongoing".
/ Completed List
Shows finished jobs. Their status will be "completed" or "failed".
Real-time Updates
The component listens for WebSocket events (batch_prediction_update on the /batch_predictions namespace) to automatically update the status, average output, and potentially move jobs between lists without requiring a page refresh.
Accessing Details
Clicking anywhere on a batch card (ongoing or completed) will trigger the opening of the details modal for that specific `batchId`.
Viewing Batch Results
Clicking on a batch card in the historical lists opens a detailed modal view (`BatchPredictionDetails.vue`) fetched via /batch-predictions/details/{batchPredictionId}:
Header Information
Provides a summary of the batch run:
- Batch Date: When the job was submitted.
- Average Output: The mean of the predicted values across the entire batch (useful for regression).
- Status: The final status (`completed`, `failed`).
- Mean Probability (Classification): If applicable, shows the average probability assigned to each class across all predictions in the batch.
Input Data Summary
Displays summary statistics (like mean, min, max, count) for the input features used in the batch prediction, provided by the `InputDataSummary` component.
Paginated Individual Results
Since batch jobs can process thousands or millions of rows, the individual results are paginated:
- Pagination Controls: Provided by `PredictionPagination.vue` to navigate through pages of results.
- Results Table: (`BatchPredictionResults.vue`) Displays a table for the current page, showing:
- Row Number (#)
- Prediction Input (Key-value pairs of features and their values for that row)
- Prediction Output (The prediction generated for that row)
Additional Details
May include a section (`batchPrediction.details`) with supplementary information or logs related to the batch run, displayed in a preformatted block.
Understanding Key Details
Here's a quick reference for some specific fields you'll encounter:
| Field | Description |
|---|---|
| Status |
|
| Average Output | The arithmetic mean of all numerical prediction outputs in the batch. Primarily useful for regression models. May be 'N/A' if not applicable or during processing. |
| Mean Probability | For classification models, this shows the average probability calculated for each class label across all predictions in the batch. Useful for understanding overall model confidence on the dataset. Format: `'ClassName1': 0.75, 'ClassName2': 0.25`. |
Related Concepts
After running batch predictions, you might explore: