|
35 | 35 | env: {GH_TOKEN: '${{ github.token }}'} |
36 | 36 | run: | |
37 | 37 | set -euo pipefail |
38 | | - curl -L --fail -H "Authorization: Bearer $GH_TOKEN" -H 'Accept: application/vnd.github+json' \ |
39 | | - https://api.github.com/repos/heathsanchez/mathgraph/actions/artifacts/9490954282/zip -o v142_artifact.zip |
| 38 | + curl -L --fail -H "Authorization: Bearer $GH_TOKEN" -H 'Accept: application/vnd.github+json' https://api.github.com/repos/heathsanchez/mathgraph/actions/artifacts/9490954282/zip -o v142_artifact.zip |
40 | 39 | mkdir -p /tmp/v142_art /tmp/v135_base |
41 | 40 | unzip -q v142_artifact.zip -d /tmp/v142_art |
42 | 41 | test -f /tmp/v142_art/trace_ace_v135_submission.zip |
|
48 | 47 | set -euo pipefail |
49 | 48 | python competitions/trace_the_ace/v144_build_smoke_probes.py --baseline /tmp/v135_base --out v144_probes |
50 | 49 | for z in v144_probes/*.zip; do |
51 | | - echo "=== $z ===" |
52 | 50 | unzip -Z1 "$z" | grep -qx 'main.py' |
53 | 51 | sha256sum "$z" |
54 | 52 | done |
@@ -82,18 +80,10 @@ jobs: |
82 | 80 | tutoringoutcomeschallengeprodacr.azurecr.io/tutoring-outcomes-runtime:gpu-latest |
83 | 81 | test -f official/submission/submission.csv |
84 | 82 | test -f official/submission/log.txt |
85 | | - python - "$tag" <<'PY' |
86 | | - import csv,sys,math,json,pathlib |
87 | | - tag=sys.argv[1] |
88 | | - with open('official/data-demo/submission_format.csv',newline='') as f: fmt=list(csv.DictReader(f)) |
89 | | - with open('official/submission/submission.csv',newline='') as f: |
90 | | - r=csv.DictReader(f); assert r.fieldnames==['response_id','probability']; out=list(r) |
91 | | - assert len(out)==len(fmt) |
92 | | - assert [x['response_id'] for x in out]==[x['response_id'] for x in fmt] |
93 | | - p=[float(x['probability']) for x in out] |
94 | | - assert all(math.isfinite(x) and 0<=x<=1 for x in p) |
95 | | - pathlib.Path('v144_evidence',tag+'.json').write_text(json.dumps({'tag':tag,'rows':len(out),'p_min':min(p),'p_max':max(p),'p_mean':sum(p)/len(p)},indent=2)) |
96 | | - PY |
| 83 | + python competitions/trace_the_ace/v144_validate_submission.py \ |
| 84 | + --format official/data-demo/submission_format.csv \ |
| 85 | + --submission official/submission/submission.csv \ |
| 86 | + --tag "$tag" --out "v144_evidence/${tag}.json" |
97 | 87 | cp official/submission/log.txt "v144_evidence/${tag}.log" |
98 | 88 | done |
99 | 89 | - uses: actions/upload-artifact@v4 |
|
0 commit comments