Improve floating point accuracy - #395

Open

avehtari wants to merge 17 commits into

improve-floating-point-accuracy

Commits on Sep 12, 2026

Commits on Sep 13, 2026

refactor: move new helper functions to helpers.R
avehtari
committed
572ae71

Commits on Sep 14, 2026

fix: handle equal infinite PSIS values
avehtari
committed
a336b76
fix: handle equal infinite stacking terms
avehtari
committed
9f4341f
fix: pass log-scale ELPD to moment matching MCSE
avehtari
committed
23005f4
fix: handle infinite log inputs
avehtari
committed
ea1cac4
fix: retain tiny positive MCSE deviations
avehtari
committed
912929c
perf: vectorize mcse_elpd and cheapen log-ratio validation
avehtari
committed
9ea4fb5
fix: propagate missing values through exp_diff_over_exp
avehtari
committed
374b8fa
fix: report -Inf log-likelihood errors in terms of the loo() input
avehtari
committed
bf22710
docs: note the normalized-weight assumption in the two-pass variances
avehtari
committed
4d884ce
style: match the surrounding blank-line spacing in the new tests
avehtari
committed
2d12c3f
test: name the importance sampling method in the -Inf log-ratio tests
avehtari
committed
32fad02