06.05.2023

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

What's New

Automatic Baselines

Automatic baselines continually update your model's reference point based on the most recently uploaded training/validation dataset. We recommend you enable auto baselines if you frequently retrain your model to streamline your monitoring workflows.
Learn more about baselines and how to enable automatic baselines here.
Enable preproduction automatic baseline

Dynamic Automatic Thresholds

Dynamically track each individual data point within your monitor to account for seasonal or slow-moving data and identify local anomalies. This is automatically enabled for all new monitors.
Autothresholds are calculated based on a statistical analysis of data over 14 days. Each day, a data point is collected, and after 14 days, the average (mean) and standard deviation of these data points are computed. The threshold is then set by adding or subtracting the standard deviation from the average.
If you're currently using the previous version of auto thresholds with existing monitors, we strongly recommend refreshing your thresholds. Learn more about thresholds here.
Example dynamic threshold in a performance monitor

❄️ Snowflake Connector

Sync model data stored in Snowflake tables with Arize to automatically stream, monitor, and analyze new model data for real-time ML observability.
Learn how to connect Snowflake with Arize here.
Connect Snowflake via Data Uplaod

Export Data To Notebook

Easily share data when you discover interesting insights so your data science team can perform further investigation or kickoff retraining workflows. Learn how to export data here.
Export data to a notebook or Phoenix

In Cluster Metrics

After choosing your desired cluster metric (e.g. euclidean distance, accuracy, custom metric, etc.), Arize automatically surfaces the clusters you should focus on for model improvement / troubleshooting so you can quickly find the root cause. Learn more here.
Select how to sort the clusters by evaluation score and dataset

Object Detection

Now in early release! Monitor and troubleshoot your Object Detection models in Arize alongside your CV, NLP, LLM, and tabular models. Sign up now for early access.
Troubleshooting Object Detection model in Arize

Enhancements

  • Prediction ID: Now optional and supports up to 128 characters. If a prediction is sent without a prediction_id, Arize will generate a prediction_id for you
  • Timestamp: The valid timestamp window has been extended from 1 year in the past to 2 years before your most recently uploaded timestamp
  • Binary Classification, Multiclass, NLP, CV Model Type: You can now log data for this model type with a score only; a label is no longer required
  • Mixed Nulls: Label and Score dataframe columns with null values can now be logged

In The News

🎓 New LLMOps Course!

Arize's newest course covers core concepts and emerging best practices for large language model operations (LLMOps). Modules include:
  • 🏗️ Foundation Models: the significance and impact of unified language models — and what lies ahead
  • 💬 Using LLMs To Evaluate LLMs: emerging techniques for assessing large language models using their peers
  • 🧠 OpenAI Eval: unlock the potential of OpenAI Evals for comprehensive and reliable assessment of LLMs
  • 🙌 LLMOps: prompt engineering and management, LLM agents, and LLM observability

📚 More Educational Content

Other new learning-focused content:

🎈Introducing: Community Paper Readings

Join us every week in the Arize Community for analysis of the latest AI research papers. Here are the first few we covered:
In this interview focused on building AI community, the founder of Cerebral Valley discusses the current moment in AI, the state of LLMOps, the mission of Cerebral Valley, and some of the exciting things community members are working on and building.