eval --compare collides on identical commands (intent twins)

Open Beginner friendly
#80 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
2/5
Estimated time
1-3 hours
Newbie friendliness
82/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Active
Tech stack
typescript

Research direction

Start in eval/run.ts at the previous-report map around line 1485, then reproduce with --corpus intent --compare using eval/corpus/intent.jsonl. Make comparison keys distinguish identical commands by evidence and label or a row id, reject duplicate keys, and verify that the comparison no longer reports twins against the wrong row.

Written by the indexing model from the issue text.

Description

eval/run.ts --compare keys the previous report by command string alone (new Map(previous.map(o => [o.command, o])), eval/run.ts:1485). When two rows share a command, the last one in the old report wins, and every current row with that command is diffed against it.

eval/corpus/intent.jsonl (#76) has 12+ twin pairs with the same command and opposite labels, so the first --corpus intent --compare run will print FIXED/REGRESSION lines against the wrong row.

Fix: key on command plus the evidence and label (or a row id), and fail loudly if two rows still share a key.

Found by the review gate on #76. It predates that PR: --corpus heldout already repeats 20 commands 25 times each.

Dominant language
TypeScript
Stars
0
Forks
1
Avg merge
1h 20m
Merged PRs (30d)
32

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from STRML/omp-classifier

All issues in STRML/omp-classifier

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.