Skip to content

Get product updates in your inbox.

What’s new in Prediction Lab.

Nov 7, 2026

Spatial smoothing and model comparison

Features

  • Fit spatially smoothed models. You can smooth geographic effects or the overall residual, choose how much smoothing to apply, and compare the baseline and smoothed fits. Spatial smoothing also works with penalized GLMs.
  • Compare model versions while you work. Compare effects and lift across versions, inspect what contributes to a selected prediction range, and see fit statistics alongside your model.
  • Define features with expressions. You can create variates and splines from expressions, control how they handle missing values, and keep your draft when switching feature types.
  • Filter lift, Lorenz, effect, and two-way actual vs predicted charts to inspect a subset of your data.
  • Write formatted descriptions across your datasets and models, including their columns and features, and link to other objects with mentions.
  • Create a dataset from a URL and see its download progress.
  • Follow a guided modeling tour using the French motor sample data.
  • Keep the AI assistant in a floating window in the desktop app.
  • Generate column descriptions with AI.
  • Organization admins can manage member roles, including read-only Viewers, in the app.

Improvements

  • Correlation calculations are faster, and distributions use less memory for columns with many categories.
  • Renames and description edits keep your existing model fits and chart results.
  • You can choose a local AI model as your default and check provider status in settings.
  • Tables scroll sideways on narrow screens, and model headers and review actions fit phone layouts.

Fixes

  • Effect charts draw the correct curves for logit, cloglog, and inverse link functions.
  • Fit-statistic comparisons exclude incompatible versions and explain what differs.
  • Deleting a feature from the model page no longer crashes the page.
  • Datetime values recorded in microseconds display correctly in the data grid.
  • Dataset uploads recover from transient database conflicts and missed completion events.