The Real-Time Prediction interface allows you to instantly generate predictions from your trained models for a single set of inputs. It provides an interactive form tailored to your model's features, enabling quick testing, exploration, and what-if analysis without writing any code.
This is ideal for situations where you need an immediate prediction based on specific, user-provided data points, such as evaluating a single customer application, testing model behavior with hypothetical inputs, or integrating manual checks into a workflow.
You typically access the real-time prediction feature from the details page of a specific trained model:
This will load the interface components specifically designed for making individual predictions.
The Real-Time Prediction screen is typically composed of several key areas:
Displays key information about the selected model, such as its name, score, type (regression/classification), training time, dataset category, and target column. This provides context for the predictions being made.
Model Name:
Customer Churn Predictor v2
Score:
0.85
Type:
Classification
Target:
Churn_Status
This is the core interactive area where you provide the input values for your prediction. Key elements include:
text
), numbers (number
with steps), and dropdowns (select
for categorical features).Displays the outcome generated by the model based on your inputs. This includes:
Provides a record of past single predictions made using this specific model via the interface. Features include:
Follow these steps to generate a real-time prediction:
Verify you are in the single/real-time prediction view. If you see options for "Batch Prediction" and "Single Prediction", ensure the latter is selected.
Fill in the values for each field in the Prediction Input Form. Use the appropriate format based on the field type (text, number, selection from dropdown).
Once all necessary inputs are provided, click the Predict button located within the Prediction Result area. The system will send your input data to the model via the /predict
API endpoint.
The Prediction Result section will update almost instantly to display the model's output based on your inputs. For classification models, class probabilities will also be shown.
The results section provides the model's output and associated confidence information:
This is the main output. For regression models, it's a continuous numerical value. For classification models, it's the most likely class label. The result briefly highlights (e.g., fades in) when updated.
Clearly indicates which variable the model is predicting (e.g., Churn_Status
).
If you are using a classification model, this section lists each possible class label along with the probability (confidence score) assigned by the model. The class label corresponding to the main `Predicted Value` is often visually distinct (e.g., bolded or colored).
Note: These probabilities represent the model's confidence in its prediction for the given input.
If feature importance data is available for your model, it will be displayed directly within the Prediction Input Form alongside each input field. This provides insight into which features had the most influence on the model's predictions *in general* (based on the training data).
A horizontal progress bar visually represents the importance score. Longer bars indicate higher importance.
The precise importance score is displayed next to the bar (e.g., `0.7521`). Higher values signify greater influence.
This global feature importance helps you understand which inputs generally drive the model's behavior. Note that this differs from Prediction Explanations, which detail the influence of inputs on a *specific* prediction.
Note: Feature importance scores are typically fetched once when the input fields are loaded via the `/generate_input_fields` endpoint.
The "Historical Predictions" component (HistoricalPredictions.vue
) keeps a log of the single predictions you've made using this interface for the current model. This is useful for reviewing past tests or analyses.
Scroll through the list to see past predictions, typically showing the date and the main prediction output.
Click the View Details button next to any prediction in the list.
A modal window will appear, displaying:
Historical data is fetched from the /predictions/example-model-123
endpoint.
Explore these related features to further leverage your models:
Understand why the model made a specific prediction.
Understand Model Decisions