Data imputation (HELM)

HELM generation task that fills one missing column in a serialized product or restaurant row, scored with quasi-exact match.

Also known as: entity_data_imputation, HELM DataImputation, DataImputation

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categoryreasoning
SubcategoryHELM missing-cell generation on Buy manufacturer and Restaurant city
Page statusunknown
Metricquasi_exact_match (schema); run spec also attaches exact-match metrics
Directionhigher_is_better
Unit%
PublisherStanford CRFM (HELM)

What it measures

entity_data_imputation is HELM's missing-cell generation scenario. The model reads a structured row serialized as "column: value" phrases with the target column stripped, then must write the missing value. HELM supports two tables from Mei et al., ICDE 2021 (IEEE 9458712): Buy, imputing manufacturer after dropping rows with empty description; and Restaurant, imputing city from the RIDDLE fz.arff file after stripping quotes and dropping NaN rows. English product and restaurant text. Open generation, not multiple choice. Laurel Orr implemented the scenario. It is not [entity_matching](entity_matching.md).

Task format

Prompt ends with the column name and a question mark (for example "manufacturer?"). HELM default adapter: generation, instructions "What is the missing value?", output noun Answer, max_train_instances 5, max_tokens 5, temperature 0. Scenario.name is entity_data_imputation; run spec names include the dataset argument.

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub