Skip to content

Release Candidate 1.2.0 - #444

Merged
EmilHvitfeldt merged 9 commits into
mainfrom
RC-1.2.0
Sep 4, 2026
Merged

EmilHvitfeldt merged 9 commits into
mainfrom
RC-1.2.0

Conversation

@EmilHvitfeldt

Copy link
Copy Markdown
Member

to close #443

The 1.2.0 win-builder submission flagged "Overall checktime > 10 min" on
Windows. Evaluating a boosted C5.0 tidypredict expression scales sharply with
trial count, and the regression test for #287 fit 18 boosted models (6 seeds
x 3 trial counts) to exercise it, accounting for 45% of the local test suite
runtime. Seeds 1 and 6 each independently reproduce the original regression
at every trial count, so the other four are dropped. Also cut the
boost_tree()/C5.0 parsnip integration test from trees = 10 to trees = 3,
since it only checks the integration plumbing, not the voting-confidence
edge case.
xrf::xrf()'s deoverlap merging cost grows sharply with rule count, and
xrf_reg_model()'s default xgb_control (nrounds = 5, max_depth = 3) generates
far more rules than needed here, making this the second-slowest test in the
suite behind the C5.0 boosting fix. A shallower control still produces
multiple rules to deoverlap and predictions still match predict(), so this
calls xrf::xrf() directly with nrounds = 3, max_depth = 2 instead of going
through the shared helper (which 11 other tests rely on with its default).
@EmilHvitfeldt
EmilHvitfeldt merged commit 56ef594 into main Sep 4, 2026
9 checks passed
@EmilHvitfeldt
EmilHvitfeldt deleted the RC-1.2.0 branch September 4, 2026 16:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Release tidypredict 1.2.0

1 participant