[TR-299] Document the use of DAAT for fast (default?) retrieval Created: 05/May/14  Updated: 16/Jun/14  Resolved: 09/Jun/14

Status: Resolved
Project: Terrier Core
Component/s: None
Affects Version/s: 3.5, 3.6
Fix Version/s: 4.0

Type: Improvement Priority: Major
Reporter: Craig Macdonald Assignee: Richard McCreadie
Resolution: Fixed  
Labels: None

Issue Links:

On ClueWeb12 segment 00, mean query response times can fall from 4.5seconds to 0.76seconds if DAAT is used instead of TAAT.

Comment by Craig Macdonald [ 05/May/14 ]

I would also suggest making taat.Full give a warning on construction if the index is more than a few million documents.

Do we need a wiki page on "making Terrier fast"

Comment by Craig Macdonald [ 06/May/14 ]

Committed change to taat package, r3793

Comment by Richard McCreadie [ 22/May/14 ]

Changed the basic default retrieval matching to daat in Manager.


Comment by Richard McCreadie [ 22/May/14 ]

Issue covered. Resolving.

Comment by Craig Macdonald [ 22/May/14 ]

Sorry, where did we agree to change the default?

Comment by Richard McCreadie [ 22/May/14 ]

I thought that we talked about this.

Most indexing tasks tackled these days will be more efficient when using daat than taat, since 'small' indices are increasingly rare.

Linking issue with TREC-365 for quantification.

Comment by Richard McCreadie [ 29/May/14 ]

From TREC-365

TAAT vs DAAT (50 queries, 300k index)

Terrier 4.0 with reverse meta + TAAT = 1.134 seconds (1.58 seconds inc. index load)
Terrier 4.0 with reverse meta + DAAT = 0.772 seconds (1.218 seconds inc. index load)

Tagging issue for resolution.

Comment by Richard McCreadie [ 09/Jun/14 ]


Generated at Thu Jan 28 01:44:43 GMT 2021 using JIRA 7.1.1#71004-sha1:d6b2c0d9b7051e9fb5e4eb8ce177ca56d91d7bd8.