Analyzing the scoring payload in Watson OpenScale
Watson OpenScale provides different methods for analyzing the scoring payloads for monitored deployments.
You can analyze the scoring payload of your model deployment by using one of the two following methods:
- Predictions by confidence: Review prediction classes and confidence distribution in each class.
- Visualize and analyze results with charts: Create a custom chart by selecting among features, prediction classes, confidence, and time.
Predictions by confidence
You can analyze the scoring payload that is sent to your deployment in the selected data range by reviewing prediction classes and confidence distribution in each class.
Visualize and analyze results with charts
To better understand model predictions and inputs at run time, use the Watson OpenScale chart builder to create custom visualizations.
The chart builder displays the output of the model’s prediction against the features or data ranges that a business considers important. For example, you can use the chart builder to see the split in predicted classes for different ranges of an attribute. You can also see confidence scores within this range of an attribute. You can analyze the scoring payload that is sent to your deployment in the selected data range by custom chart, selecting between features, prediction classes, and confidence.
It helps uncover new trends in the data, which can prompt the business and data science teams to consider changes to the AI model.
Next steps
Related topics: Read a blog post about Model Behavioural Insights using IBM Watson OpenScale
Parent topic: Getting insights with Watson OpenScale