Buckets:
| # cwm3/eval/run_joint_evals_v2.sh — paired comparisons for the pack-v2 joint runs (dirs I1_<variant>_s<k>_v2; tags I1v2_s<k>_<a>_vs_<b>), plus | |
| # the v2-vs-v1 pair of the same variant on identical rows (tag I1v2_s<k>_<a>_vs_<a>v1), idempotent. bash eval/run_joint_evals_v2.sh "0 1 2" | |
| 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() { # tag a_dir b_dir | |
| 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" | |
| done | |
| } | |
| for s in ${1:-0 1 2}; do | |
| pair "I1v2_s${s}_all_vs_indep" "I1_all_s${s}_v2" "I1_indep_s${s}_v2" | |
| pair "I1v2_s${s}_indep_vs_indepv1" "I1_indep_s${s}_v2" "I1_indep_s${s}" | |
| pair "I1v2_s${s}_all_vs_allv1" "I1_all_s${s}_v2" "I1_all_s${s}" | |
| done | |
| $PY eval/report_joint.py --stage I1v2 --a all --b indep --claim B2 --seeds ${1// /,} --extra indep:indepv1 all:allv1 --out results/tables/stage_I1_v2.md 2>&1 | grep -E "written|Claim" | |
| echo I1V2_EVALS_DONE | |
Xet Storage Details
- Size:
- 1.45 kB
- Xet hash:
- f8c52b40c96cd0e6f000a8c372d8fc708c6e6aa10550f8ca51bfc7164266d4bd
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.