Lower latencies and new models for HPO

Myst Platform Release (2022-08-19)

Hello Myst Platform users,

We’ve got a few exciting improvements for you this week!

Please update your myst-alpha package at your earliest convenience to ensure you have access to this week’s updates (instructions here).

⚡️ Enhancements:

• One of our top priorities this quarter is to generate predictions faster on the platform. We’ve made a few improvements in this release that have reduced median prediction latency from 40-50%. We will continue iterating towards even lower latencies, particularly at the top of the hour, but wanted to share these interim results.
• You can now use XGBoost, LightGBM, and Elastic Net with any objective function in an HPO. This is a starting point for our HPO feature. If there are other models you’re keen to use in an HPO, please let us know.
  Note: When you run myst.hpo.Hyperopt you no longer have to specify the metric parameter. We now infer this for you based on the model and objective function you have selected. If you have been using the client library to run HPOs, remove this parameter before running your next HPO.

✨ New feature:

• Logistic Regression is live! You can use this model to estimate the probability of RT prices exceeding DA prices, for example. Learn more about this model here.

📘

A tip!

If your models are fitting every hour, consider updating them to fit once per day. Through our own forecasting service work, we have found that daily fits deliver comparable accuracy to hourly fits. Additionally, when we transition to usage-based costs, we envision that most users will prefer daily fits to hourly fits given the tradeoff between cost and marginal accuracy improvements.

As always, please do not hesitate to reach out via the chat embedded in the Myst Platform or email us at [email protected] with questions or feedback.

Thank you!
Ellery and the Myst team