Read the fitted models' objective with xgb.config()

[eliotmcintire]

Every prediction run (hindcasts and NRV) stopped at the ignition step, because the fitted models' objective was read the old way and the current xgboost no longer stores it there. The objective is now read with xgboost's own function.

Affects: prediction runs with xgboost 3.x fitted models now run; results are unchanged.

Technical details

Defect: fitObjective() (fireSense_ignitionPredict.R:275) read model$params$objective; a reloaded xgboost 3.x booster is a pointer object with no $params, so it stopped with "cannot read the objective". It now uses xgboost::xgb.config(model)$learner$objective$name, falling back to attr(model, "params")$objective, and keeps the error (message now names both). xgboost (>= 3.0.0) added to reqdPkgs. No other xgboost internals are read in this module or in fireSense_burn development.

Tests: new test-xgboost-objective.R trains real boosters, round-trips them through saveRDS/readRDS, and checks the objective and the mixed-objective error; it fails on development and passes here. The toy fold stand-ins now keep the objective in attr(, "params"), as real models do. Module tests via convertToPackage() + test_local(): development plus the new test 35 pass / 2 fail; this branch 37 pass / 0 fail.

🤖 Generated with Claude Code

https://claude.ai/code/session_014G18Mik3fH4z5M5CyMSUS6