experimental/cuda-ubi9/: rouge-score-0.1.2 metadata and description

Homepage Simple index

Pure python implementation of ROUGE-1.5.5.

author Google LLC
author_email rouge-opensource@google.com
  • Programming Language :: Python :: 3
  • License :: OSI Approved :: Apache Software License
  • Operating System :: OS Independent
description_content_type text/markdown
  • absl-py
  • nltk
  • numpy
  • six >=1.14.0
requires_python >=3.7
File Tox results History
24 KB
Python Wheel

Python ROUGE Implementation


This is a native python implementation of ROUGE, designed to replicate results from the original perl package.

Maintainers may be contacted at rouge-opensource@google.com.

ROUGE was originally introduced in the paper:

Lin, Chin-Yew. ROUGE: a Package for Automatic Evaluation of Summaries. In Proceedings of the Workshop on Text Summarization Branches Out (WAS 2004), Barcelona, Spain, July 25 - 26, 2004.

ROUGE for Python

There are ROUGE implementations available for Python, however some are not native python due to their dependency on the perl script, and others provide differing results when compared with the original implementation. This makes it difficult to directly compare with known results.

This package is designed to replicate perl results. It implements:

Note that not all options provided by the original perl ROUGE script are supported, but the subset of options that are implemented should replicate the original functionality.

Stopword removal

The original ROUGE perl script implemented optional stopword removal (using the -s parameter). However, there were ~600 stopwords used by ROUGE, borrowed from another now defunct package. This word list contained many words that may not be suited to some tasks, such as day and month names and numbers. It also has no clear license for redistribution. Since we are unable to replicate this functionality precisely we do not include stopword removal.

Two flavors of ROUGE-L

In the ROUGE paper, two flavors of ROUGE are described:

  1. sentence-level: Compute longest common subsequence (LCS) between two pieces of text. Newlines are ignored. This is called rougeL in this package.
  2. summary-level: Newlines in the text are interpreted as sentence boundaries, and the LCS is computed between each pair of reference and candidate sentences, and something called union-LCS is computed. This is called rougeLsum in this package. This is the ROUGE-L reported in Get To The Point: Summarization with Pointer-Generator Networks, for example. If your references/candidates do not have newline delimiters, you can use the --split_summaries flag (or optional argument in RougeScorer).

How to run

This package compares target files (containing one example per line) with prediction files in the same format. It can be launched as follows (from google-research/):

python -m rouge.rouge \
    --target_filepattern=*.targets \
    --prediction_filepattern=*.decodes \
    --output_filename=scores.csv \
    --use_stemmer=true \

Using pip

pip install -r rouge/requirements.txt
pip install rouge-score

Then in python:

from rouge_score import rouge_scorer

scorer = rouge_scorer.RougeScorer(['rouge1', 'rougeL'], use_stemmer=True)
scores = scorer.score('The quick brown fox jumps over the lazy dog',
                      'The quick brown dog jumps on the log.')


Licensed under the Apache 2.0 License.


This is not an official Google product.