Model Evaluation

Focus on the problems that matter

It’s critical to understand where your model is performing badly, diagnose why the mistakes are happening, and triage the most important problems to solve first.

Analyze model accuracy metrics

  • Automatically calculate accuracy metrics for a variety of machine learning tasks.
  • Analyze metrics at a high level, then zoom in and visualize the actual datapoints where the errors occurred.

Find patterns of failures

Visualize model failures grouped by similarity instead of manually digging through your dataset.
Quickly find systematic problems in your data or model performance and decide on the right action for resolution.

Compare your models

  • View similarities and differences in models across a standardized set of data and accuracy metrics.
  • Compare release candidates to production models. View metrics comparisons at a high level, then zoom into the exact examples where performance is improving or regressing.

Keep track of failures

  • Group failure cases together with Segments and collaborate with your team to resolve them.
  • Set up machine learning regression tests to track model accuracy on critical failures as you train new models.
Never regress your model performance. Ensure your model is improving overall and on critical segments of your dataset.

Get in touch

Schedule time to get started with Aquarium