Files
agent/apps/keypoint-eval/README.md
Anibal Angulo a53f8fcf62 First commit
2026-02-18 19:57:43 +00:00

1.0 KiB

Keypoint Evaluator

This application evaluates a RAG (Retrieval-Augmented Generation) system based on the keypoint methodology from the RAGEval paper.

How to use

To run the evaluation, execute the following command from the root directory of the project:

python -m keypoint_eval.main --evaluation-name <EVALUATION_NAME> --matriz-eval <PATH_TO_EVALUATION_MATRIX_FILE>

Arguments

  • --evaluation-name: The name of the evaluation.
  • --matriz-eval: The path to the evaluation matrix file.

The application will read the evaluation matrix from the specified file and will generate a CSV and a JSON file with the evaluation results.

Input File Structure

The input file can be a CSV, Excel, or JSON file.

The file must contain the following columns:

  • input: The user's question.
  • expected_output: The ground truth or expected answer.
  • category (optional): The category of the question.

If the input column is not found, the application will look for columns containing "pregunta" or "question".