0.7.0

PyTerrier 0.7.0

Notable update: containing a small number of new features, and numerous other minor updates. New features include new operators on Terrier indices (#196), as well as the ability to express a Terrier weighting model in a Python lambda (#208, #215). A number of upstream fixes in terrier-core supported multi-lingual settings, gridsearch etc. There was also enchanced support for ir_measures

• #227, #230 dataframe type coercion during evaluation
• #226, #228 pt.Experiment consistency in missing qids in topics/qrels/runs
• #224 Using PyTerrier when not connected to Internet
• #220, #221 pt.debug utility transformers
• #219 pt.apply.new_column fails on empty dataframes
• #216 Mirrors for datasets
• #208, #215 Terrier weighting models expressed in Python
• #211 only compute ranks for score column
• #209 pt.apply.by_query verbose support
• #202, #231 pandas 1.3.0 compatibility
• #201 ParallelIndexer should use indexer.setExternalParalllism()
• #200 BatchRetrieve.from_dataset() support for http://data.terrier.org/
• #197 meaningful error message when passing corpus_iter to TRECCollectionIndexer
• #196 add + and len() support on a Terrier index.
• #195 intersect operator returns extra columns
• #192 more documentation on how to tune weightmodels
• #191 Revised GridSearch documentation
• #188 Global properties for BatchRetrieve
• #187 dataset indices with different variants dont get different dirs
• #185 Addititional dataset.get_topicsqrels() to make pt.Experiment faster to write
• #178 rank accuracy under negative scalar multiple
• #177 add a dataset search feature
• #176 Evaluating empty resultset
• #136 more pre-built indices?
• #113 datasets to support mirrors
• #21, #222 Alias BatchRetrieve to TerrierRetrieve

0.6.0

Significant update: Windows support, new evaluation measures, GridScan/Search, multi-threaded retrieval for Terrier's BatchRetrieve.

• Windows support - #135
• Use ir_measures for calculating measures -- thanks to Sean MacAvaney, University of Glasgow
• GridScan and GridSearch - #97 -- thanks to Chentao Xu, University of Glasgow
• Conducting experiments in batches of queries, and dropping unused queries - #98, #81
• Keep previous formulations of the query - #126
• Terrier retrieval using multiple threads - #3
• Retrieve more than 1000 results from Terrier - #140, thanks to @DayalStrub
• Update pandas version in requirements.txt (#159), thanks to Alberto Ueda, UFMG
• Fixes to FilesIndexer, thanks to Chirag Shag, University of Washington
  plus other minor updates

0.5.0

This release includes:

• multiple testing correction in pt.Experiment() (#99)
• Terrier Index API needs documentation (#103)
• Tokenising/escaping queries from IR Datasets (#104), from Sean MacAvaney, University of Glasgow
• BatchRetrieve should warn and ignore empty queries (#110), with thanks to @TimMo-prog
• Indexing pipelines (#118)
• pt.apply operations for making a new column and dropping a column (#121), with thanks to Sean MacAvaney, University of Glasgow
• detect LTR learned model use that differs in number of features from fitting time (#123), with thanks to Iadh Ounis, University of Glasgow

There are also now many additional neural ranking and re-ranking plugins for PyTerrier - see list at https://pyterrier.readthedocs.io/en/latest/neural.html#available-neural-re-ranking-integrations

0.4.0

This release includes:

• IR Datasets support (#90, #100) - thanks to Sean MacAvaney, University of Glasgow
• faster indexing via multi-threading and fifos (#92) - thanks to Sean MacAvaney, University of Glasgow
• Easier ways of operating on text of documents (#88)
• Documentation improvements - thanks to Iadh Ounis, University of Glasgow
• ensuring Precision@cutoff metrics are reported (#89) - thanks to Xiao Wang, University of Glasgow
• rounding of pt.Experiment values - suggestion by Iadh Ounis, University of Glasgow

0.3.1

This PyTerrier release contains significant updates from the last public release in March 2020. It is intended to serve as a stable release for PyTerrier, while further improvements are merged in a new stable release.