Skip to content

Correct disposable card limit label - #15

Open
johnquevedo wants to merge 1 commit into
PolyAI-LDN:masterfrom
johnquevedo:agent/correct-disposable-card-limit
Open

Correct disposable card limit label#15
johnquevedo wants to merge 1 commit into
PolyAI-LDN:masterfrom
johnquevedo:agent/correct-disposable-card-limit

Conversation

@johnquevedo

Copy link
Copy Markdown

Summary

Correct one BANKING77 test label:

  • text: how many transactions can i make with a disposable card
  • previous label: get_disposable_virtual_card
  • proposed label: disposable_card_limits
  • location: banking_data/test.csv, zero-based data row 2,637

This PR changes no other dataset row.

Evidence

The source was pinned to publisher commit
57ec275d8078af65b7731c2a98be812d844a6d6b, which is also the current
master commit at the time of this PR.

The wording asks about the number of permitted transactions. Closely matching
training rows are labeled disposable_card_limits:

  • train row 4,393: “How many transactions can I do with my disposable card?”
  • train row 4,405: “How many transactions can I make on my disposable card?”
  • train row 4,420: “How many transactions can I do with a disposable card?”

Reproduction

From the repository root:

python - <<'PY'
import csv
from pathlib import Path

test = list(csv.DictReader(Path("banking_data/test.csv").open()))
train = list(csv.DictReader(Path("banking_data/train.csv").open()))

print(len(test), test[2637])
for index in (4393, 4405, 4420):
    print(index, train[index])
PY

After this patch, the test file still parses to 3,080 rows and row 2,637 has
category disposable_card_limits.

Limitations

The repository does not publish a category annotation guide, so this proposal
relies on the utterance meaning and consistent labels of near-identical
training examples. Automated nearest-neighbor and character n-gram models
helped surface and cross-check the candidate, but their outputs are supporting
signals rather than proof. This PR intentionally excludes other uncertain
findings and makes no claim of significant downstream model improvement.

@johnquevedo
johnquevedo marked this pull request as ready for review July 25, 2026 20:26
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant