[
https://issues.apache.org/jira/browse/OPENNLP-1330?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17363601#comment-17363601
]
Jeff Zemerick commented on OPENNLP-1330:
----------------------------------------
Hi [~eihli], thanks a lot for reporting this. Do you mind opening a pull
request for the change at [https://github.com/apache/opennlp?] I can do it for
you from your patch file if you like but I prefer for you to get credit for
your fix.
> Parser top k parses doesn't show "top" (highest probability) parses.
> --------------------------------------------------------------------
>
> Key: OPENNLP-1330
> URL: https://issues.apache.org/jira/browse/OPENNLP-1330
> Project: OpenNLP
> Issue Type: Bug
> Components: Parser
> Affects Versions: 1.9.4
> Reporter: Eric Ihli
> Priority: Minor
> Labels: Parser
> Attachments: FixParserTopK.patch
>
> Original Estimate: 0h
> Remaining Estimate: 0h
>
> The default "top n parses" of 1 shows the most likely parse; one with a log
> prob of -2.9.
> Passing in anything greater than 1 returns the least probable parses.
> {code:java}
> ➜ bin git:(master) echo "Eric is testing." | ./opennlp Parser -k 1
> ~/src/prhyme/resources/models/en-parser-chunking.bin
> Loading Parser model ... done (2.515s)
> 0 -2.913239374955309 (TOP (S (NP (NNP Eric)) (VP (VBZ is)) (. testing.)))
> Average: 41.7 sent/s
> Total: 1 sent
> Runtime: 0.024s
> Execution time: 2.585 seconds
> ➜ bin git:(master) echo "Eric is testing." | ./opennlp Parser -k 2
> ~/src/prhyme/resources/models/en-parser-chunking.bin
> Loading Parser model ... done (2.578s)
> 0 -14.968552218782305 (TOP (S (NP (NNP Eric)) (SBAR (S (ADVP (VBZ is)) (VBG
> testing.)))))
> 1 -13.957766393572408 (TOP (S (SBAR (S (NP (NNP Eric)) (ADVP (VBZ is)))) (VBG
> testing.)))
> Average: 95.2 sent/s
> Total: 2 sent
> Runtime: 0.021s
> Execution time: 2.640 seconds
> {code}
> The fix is clear and simple. We should be taking from the first of the
> TreeSet rather than from the end.
> {code:java}
> else if (numParses == 1) {
> return new Parse[] {completeParses.first()};
> }
> else {
> List<Parse> topParses = new ArrayList<>(numParses);
> while (!completeParses.isEmpty() && topParses.size() < numParses) {
> Parse tp = completeParses.last(); //// <--- Change to .first()
> completeParses.remove(tp);
> topParses.add(tp);
> //parses.remove(tp);
> }
> return topParses.toArray(new Parse[topParses.size()]);
> }{code}
> After patch, results are what I expect.
>
> {code:java}
> ➜ bin git:(master) ✗ echo "Eric is testing." | ./opennlp Parser -k 1
> ~/src/prhyme/resources/models/en-parser-chunking.bin
> Loading Parser model ... done (2.517s)
> 0 -2.913239374955309 (TOP (S (NP (NNP Eric)) (VP (VBZ is)) (. testing.)))
> Average: 45.5 sent/s
> Total: 1 sent
> Runtime: 0.022s
> Execution time: 2.580 seconds
> ➜ bin git:(master) ✗ echo "Eric is testing." | ./opennlp Parser -k 2
> ~/src/prhyme/resources/models/en-parser-chunking.bin
> Loading Parser model ... done (2.530s)
> 0 -2.913239374955309 (TOP (S (NP (NNP Eric)) (VP (VBZ is)) (. testing.)))
> 1 -3.1132674983145825 (TOP (S (NP (NNP Eric)) (VP (VBZ is) (NP (NN
> testing.)))))
> Average: 90.9 sent/s
> Total: 2 sent
> Runtime: 0.022s
> Execution time: 2.596 seconds
> {code}
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
