alpha-nDCG

Python implementation of alpha nDCG, tested with ndeval (test program code is provided, except no files to try them out).

Usage

Single query, returns score

from alpha_nDCG import AlphaNDCG
alpha_nDCG = AlphaNDCG('ndeval-format-qrel')
alpha_nDCG.compute(['docid1', 'docid2', 'docid3'], 'qid')

Multiple queries, returns dict of scores

alpha_nDCG.compute_multi([
	(['docid1', 'docid2', 'docid3'], 'qid1'),
	(['docid1', 'docid2', 'docid3'], 'qid2')
])

By default computes alpha-nDCG@20. To change depth:

alpha_nDCG.compute(['docid1', 'docid2', 'docid3'], 'qid', depth=20)

References

Clarke, Charles LA, et al. "Novelty and diversity in information retrieval evaluation." Proceedings of the 31st annual international ACM SIGIR conference on Research and development in information retrieval. ACM, 2008. [pdf]