QryEval consists of the following classes.

• Access to the index

  • Idx: Manage and provide access to a Lucene index.
  • PyLu: Low-level management and access to a Lucene index through Java. Used by Idx.

• Essential data structures

  • InvList: Create, access, and manipulate inverted lists.
  • Results: Create, access, and manipulate a list of search results.
  • ScoreList: Create, access, and manipulate document score lists.

• DaaT Query evaluation

  • Qry: The Qry class is the root class in the query operator hierarchy.
  • QryIop: All query operators that return inverted lists are subclasses of the QryIop class.
  • QryIopSyn: The SYN operator for all retrieval models.
  • QryIopTerm: The TERM operator for all retrieval models.
  • QryParser: A simple query parser.
  • QrySop: QrySop is the root class of all query operators that calculate a score for a document.
  • QrySopOr: The OR operator for all retrieval models.
  • QrySopScore: The SCORE operator for all retrieval models.
  • RetrievalModel: RetrievalModel is the root class in the retrieval model hierarchy.
  • RetrievalModelUnrankedBoolean: Define and store data for the Unranked Boolean Retrieval Model.

• Ranking pipeline

  • Ranker: Get an initial ranking for a set of queries.
  • Reranker: Rerank an initial ranking (HW2, HW4).
  • RerankerWithLtr: Use feature-based learning to rank to rerank an initial ranking (HW2).
• Utilities
  • TeIn: Write search results in trec_eval format.
  • Timer: A simple timer.
  • Util: Handy utilities.

For Your Information

The Java libraries are accessed using pyjnius. QryEval can also use jpype. Both produce the same results. In our initial testing, we found pyjnius to be a little faster than jpype, so QryEval defaults to pyjnius, however both are supported.

Although pyjnius and jpype are amazing, both are quite slow for certain kinds of index access, so QryEval also uses its own Java helper class (IdxHelper.jar).