language-evaluation (Experimental)

Collection of evaluation code for natural language generation.

Note: API could change frequently without notice

Metrics

CocoEvaluator: coco-caption (BLEU1-4, METEOR, ROUGE, CIDEr, SPICE)
RougeEvaluator: sentence-level rouge (ROUGE-1, ROUGE-2, ROUGE-L with f-measure)
Rouge155Evaluator: summary-level rouge (ROUGE-1, ROUGE-2, ROUGE-L with f-measure)

Requirements

Java 1.8.0+ (used by coco-caption evaluator)
Python 3.6+
libxml-parser-perl (used by ROUGE.1.5.5.pl)

Installation and Usage

Install external dependencies (e.g. Java 1.8.0+, libxml-parser-perl):

# Oracle Java
sudo add-apt-repository ppa:webupd8team/java
sudo apt upadte
apt-get install oracle-java8-installer

# libxml-parser-perl
sudo apt install libxml-parser-perl

Then run:

pip install git+https://github.com/bckim92/language-evaluation.git
python -c "import language_evaluation; language_evaluation.download('coco')"

Python API (or see language_evaluation_test.py):

import language_evaluation
from pprint import PrettyPrinter
pprint = PrettyPrinter().pprint

predicts = ['i am a boy', 'she is a girl']
answers = ['am i a boy ?', 'is she a girl ?']

evaluator = language_evaluation.CocoEvaluator()
results = evaluator.run_evaluation(predicts, answers)
pprint(results)
# {'Bleu_1': 0.9999999997500004,
#  'Bleu_2': 0.5773502690332603,
#  'Bleu_3': 4.3679023223468616e-06,
#  'Bleu_4': 1.4287202142987477e-08,
#  'CIDEr': 3.333333333333333,
#  'METEOR': 0.43354749322305886,
#  'ROUGE_L': 0.75,
#  'SPICE': 0.6666666666666666}

evaluator = language_evaluation.RougeEvaluator(num_parallel_calls=5)
results = evaluator.run_evaluation(predicts, answers)
pprint(results)
# {'rouge1': 1.0,
#  'rouge2': 0.3333333333333333,
#  'rougeL': 0.75}

evaluator = language_evaluation.Rouge155Evaluator(num_parallel_calls=5)
results = evaluator.run_evaluation(predicts, answers)
pprint(results)
# {'rouge1': 1.0,
#  'rouge2': 0.3333333333333333,
#  'rougeL': 0.75}

Notes

TODOs
- Support more metrics (e.g. embedding-based)
- Support command-line interface
- Support full functionality and configuration for rouge
- Implement summary-level rouge scorer in pure python
- Add tests & CI

Related Projects

License

See LICENSE.md.