New Releases, Enhancements, Changes + Arize in the News!

What's New

Alerting Integrations: PagerDuty & OpsGenie

Our native integrations with your alerting tools streamline your monitoring workflows. Use integrations to:

  • Tailor alerts to specific model dimensions and metrics that matter the most

  • Send comprehensive metadata via your incident management flow to catch and debug your model issues faster

  • Store integration keys at the organizational level for easy setup and system organization

Learn how to use our PagerDuty integration here and learn how to use our OpsGenie integration here.

Comparison Dataset in the Drift Tab and Dimension Details

Compare your current drift against your model baseline and custom baseline even before creating monitors. Find this new addition in the 'Drift' Tab, or in the Feature Details page in your 'Model Breakdown' card on the 'Performance Tracing' page.

Learn more about model baselines here.

Automatic Thresholds on Data Quality Monitors

You can now enable auto thresholds on custom and managed data quality monitors. Enable or disable automatic thresholds for data quality monitors in the UI or programmatically with our public API.

Learn more about automatic thresholds and how to customize thresholds here.

Learn more about our public API here and how to programmatically edit monitors here.

Bulk Monitor Creation

Use the new bulk monitor creation flow to automatically set up performance, drift, and data quality monitors at a click of a button. Choose from various performance metrics, set your positive class, and add an alerting integration all in the same workflow.


UMAP Enhancements

Select Group of Embeddings with Lasso Tool

Easily grab a group of embeddings on the UMAP to see more info and troubleshoot by selecting the lasso tool.

New "Color By" Options For Embeddings Analysis

Additional options have been added to the "Color By" dropdown when visualizing embeddings on the UMAP plot for faster analysis and discovering problematic segments. Choose between:

  • Dataset

  • Prediction Label

  • Actual Label

  • Correctness (whether or not the prediction label matches the actual label)

In the News

Visualizing Your Embeddings: An Evolutionary Guide from SNE to t-SNE to UMAP

A digestible guide to understanding the underlying logic and evolution from SNE to t-SNE to UMAP presented by Arize Data Scientist, Kiko Castillo.

Four Takeaways from Arize:Observe Unstructured

Arize:Observe Unstructured wrapped up in June before a large audience of technical leaders and ML practitioners. Did you miss the event? Here are four top takeaways:

  1. Monitoring of Unstructured Data Has Arrived!

  2. What Works In Training May Not Work In Production When Deciding What To Label Next

  3. The Rise of the Single, Unified Model Will Change MLOps

  4. Cutting-Edge Machine Learning Is Becoming More Accessible

Read more for context from Hugging Face, OpenAI and others and watch on-demand sessions here.

Three Pitfalls To Avoid With Embeddings

There are a few gotcha moments with embeddings. Aparna Dhinakaran, CPO of Arize AI, offers practical advice in three areas:

  1. How to version your project

  2. How to monitor your embedding once it goes live, and

  3. How to get an intuitive sense for the quality of your embeddings.

Interview: Justin Chen of Google

Justin Chen, Software Engineer (ML) at Google, details his career arc and shares best practices for productionizing ML models in a wide-ranging interview with Arize ML Engineer, Amber Roberts.

Last updated