v2.5.0

Highlights

• Baseline models in backtests: Model templates can now take a baseline or comparison role. New seasonal median, persistence and global median baselines run once per backtest specification, and their scores appear next to the evaluated model on the leaderboard.
• MLproject models as chapkit services: --run-config.as-chapkit runs MLproject models as chapkit services (requires chapkit 2.4.0). See the MLproject tutorial.

Features

• Add a baseline/comparison role to model templates (#712)
• Add seasonal median and persistence baselines (CLIM-1344) (#713)
• Add built-in Python models and a global median baseline (#714)
• Run the baseline models once per backtest specification (#724)
• Return baseline and comparison model scores on the leaderboard (CLIM-1347) (#735)
• Raise pending alerts from stored predictions (CLIM-1335) (#702)
• Add a simulate entry point and chap simulate command for simulation models (reference) (#716)
• Run MLproject models as chapkit services with --run-config.as-chapkit (CLIM-1352) (#709, #737)
• Support DHIS2 ApiToken authentication (CLIM-1424) (#728)
• Add comparison operators to metric specs (#708)
• Read target and hpo_search_space from chapkit service info (#707)

Bug Fixes

• Forecast the backtest target in prediction setups (CLIM-1341) (#706)
• Send an empty covariate list to chapkit services instead of dropping it (#726)
• Stop service registry reads from stalling when Redis is down (#723)
• Stop the Redis availability probe from retrying a refused connection (#718)
• Replace deprecated Redis setex in the orchestrator (#719)
• Stop _is_chapkit_url from swallowing unexpected exceptions (#721)

Documentation

• Tutorial for building a chapkit model from an MLproject file (#736)
• Fix wrong commands, versions and links found in the docs review (#743)

Full Changelog: v2.4.0...v2.5.0