Implementation:Explodinggradients Ragas DataCompyScore Metric

Field	Value
source	Repo
domains	Metrics, Data_Comparison
last_updated	2026-02-10

Overview

DataCompyScore compares CSV-formatted data in a reference and response using the datacompy library, computing precision, recall, or F1 at the row or column level.

Description

The DataCompyScore class evaluates the quality of generated tabular data by parsing both the reference and response as CSV strings, then comparing them using the datacompy.Compare library. It supports two comparison modes (rows vs columns) and three scoring metrics (precision, recall, F1). The metric does not require an LLM and inherits only from SingleTurnMetric.

Key attributes:

• mode -- Comparison granularity: "rows" (default) or "columns".
• metric -- Scoring type: "precision", "recall", or "f1" (default "f1").

Dependencies: Requires pandas and datacompy packages.

Usage

The metric requires reference and response columns, both as CSV-formatted strings.

Code Reference

Property	Value
Source Location	src/ragas/metrics/_datacompy_score.py L16-79
Class Signature	class DataCompyScore(SingleTurnMetric)
Import	from ragas.metrics import DataCompyScore

I/O Contract

Inputs

Parameter	Type	Required	Description
reference	str	Yes	CSV-formatted string of the ground truth data
response	str	Yes	CSV-formatted string of the generated data

Outputs

Output	Type	Description
score	float	Comparison score (0.0 to 1.0) based on the selected mode and metric, or NaN on parsing error

Usage Examples

from ragas.metrics import DataCompyScore
from ragas.dataset_schema import SingleTurnSample

metric = DataCompyScore(mode="rows", metric="f1")

sample = SingleTurnSample(
    reference="name,age\nAlice,30\nBob,25",
    response="name,age\nAlice,30\nBob,25\nCharlie,35"
)
# score = await metric.single_turn_ascore(sample)

Related Pages

• Explodinggradients_Ragas_BleuScore_Metric -- Text-level comparison metric
• Explodinggradients_Ragas_StringMetrics_Module -- String-level comparison metrics
• Explodinggradients_Ragas_FactualCorrectness_Metric -- Factual accuracy evaluation for text