AsianPlayer/TriDrive / code /eval /run_joint_evals_v2b.sh
AsianPlayer's picture
download
raw
1.89 kB
#!/usr/bin/env bash
# cwm3/eval/run_joint_evals_v2b.sh — B2 re-adjudication on the CORRECTED v2 joint runs (2026-09-11): fresh tags I1v2b_s<k>_* (the I1v2_* rows were computed from the
# first-attempt INVALID layouts and are left untouched). Pairs per seed (identical rows, paired_compare_joint):
# all_vs_indep (B2 baseline), allminus_vs_indep (candidate survivor `all_minus_road2drv`), allminus_vs_all, indep_vs_indepv1, all_vs_allv1.
# bash eval/run_joint_evals_v2b.sh "0 1 2" [render=1]
set -u
cd "$(dirname "$0")/.."
PY=python
export OMP_NUM_THREADS=4 MKL_NUM_THREADS=4
have_tag() { grep -q ",$1,$2,test," results/paired_compare.csv 2>/dev/null; }
pair() { local tag="$1" a="$2" b="$3"
[[ -f results/runs/$a/PRED_DONE && -f results/runs/$b/PRED_DONE && -f results/runs/$a/TRAIN_DONE && -f results/runs/$b/TRAIN_DONE ]] || { echo "skip $tag (runs not finished: TRAIN_DONE + PRED_DONE required)"; return; }
for reg in manual engaged; do have_tag "$tag" "$reg" && continue; echo "== $tag $reg"
$PY eval/paired_compare_joint.py --regime "$reg" --a "results/runs/$a" --b "results/runs/$b" --tag "$tag" 2>&1 | grep -E "appended|Error|Traceback|File \""; done; }
for s in ${1:-0 1 2}; do
pair "I1v2b_s${s}_all_vs_indep" "I1_all_s${s}_v2" "I1_indep_s${s}_v2"
pair "I1v2b_s${s}_allminus_vs_indep" "I1_all_minus_road2drv_s${s}_v2" "I1_indep_s${s}_v2"
pair "I1v2b_s${s}_allminus_vs_all" "I1_all_minus_road2drv_s${s}_v2" "I1_all_s${s}_v2"
pair "I1v2b_s${s}_indep_vs_indepv1" "I1_indep_s${s}_v2" "I1_indep_s${s}"
pair "I1v2b_s${s}_all_vs_allv1" "I1_all_s${s}_v2" "I1_all_s${s}"
done
if [[ "${2:-1}" == "1" ]]; then S=${1:-0 1 2}
$PY eval/report_joint.py --stage I1v2b --a allminus --b indep --claim B2 --seeds ${S// /,} --extra all:indep allminus:all indep:indepv1 all:allv1 --out results/tables/stage_I1_v2b.md 2>&1 | grep -E "written|verdict"
fi
echo I1V2B_EVALS_DONE

Xet Storage Details

Size:
1.89 kB
·
Xet hash:
94805953b479df92458092d270b3455cb0eac8aba1fc1c86226e1d8bdc151729

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.