Map columns in datasets

Map columns in datasets to different names. This helps you align the column names in your dataset with the column names expected by the scorer.

Python

In the Weave Python SDK, use the column_map attribute on a scorer to map its expected parameter names to your dataset’s column names. The mapping format is {scorer_parameter: dataset_column}.

The following example maps the output and target parameters to a dataset’s model_output and answer columns:

from weave.scorers import EmbeddingSimilarityScorer

similarity_scorer = EmbeddingSimilarityScorer()

similarity_scorer.column_map = {
    "output": "model_output",  # The model's generated text
    "target": "answer"         # The expected or reference response
}

For more details on scorer column mapping, see Mapping Column Names with column_map.

TypeScript

In the Weave TypeScript SDK, column mapping is configured on the Evaluation object using the columnMapping option, not on individual scorers. The mapping format is {scorer_key: dataset_column}.

The following example maps expectedOutputTimesTwo (used by the scorer) to the expected column in the dataset:

import * as weave from 'weave';

const myScorer = weave.op(
  ({modelOutput, datasetRow}) => {
    return modelOutput * 2 === datasetRow.expectedOutputTimesTwo;
  },
  {name: 'myScorer'}
);

const myEval = new weave.Evaluation({
  id: 'my-evaluation',
  dataset: [{expected: 2}],
  scorers: [myScorer],
  columnMapping: {expectedOutputTimesTwo: 'expected'},
});

For more details on TypeScript scorer arguments, see Scorer Keyword Arguments.