Commit Graph

20 Commits (5ae1d3593e72f49c710c7cf69132a898b5f9bd90)

Author SHA1 Message Date
Ken Wakasa de8a9a8227 Small cleanups
Change-Id: Ib66507b8934bc8019a762d24d5311411e044ec84
2012-08-17 13:06:28 +09:00
Jean Chalard b14fc88e48 Tag the whitelisted entries in native code.
Since this is already used in Java land, this actually does
activate the whitelist path, and the code is now fully
functional. We still have to remove the old whitelist resource
and to compile the dictionary that includes the whitelist.

Bug: 6906525
Change-Id: Iacde5313e303b9ed792940efaf6bcfa4ee1317bd
2012-08-13 16:35:59 +09:00
Ken Wakasa 77e8e81ad9 Header cleanup. Moved a couple of functions from .h to .cpp.
Change-Id: Ifd12a7632f75395bd0ef5e394d5c2abd6cbe28c6
2012-08-02 20:19:39 +09:00
Jean Chalard 195605084e Move flags belonging to BinaryFormat to the right place.
These masks and flags are constants that are an integral part
of the format. They belong in BinaryFormat and have nothing to
do in UnigramDictionary.
This needs I6751dda4 to not break the build

Bug: 6429243
Change-Id: Ic1c842b3245f7fdc25aa8d1459c5bb07b262e265
2012-08-01 00:23:52 +09:00
Ken Wakasa 0bbb917d12 Cosmetic fixes and style fixes
Change-Id: I69c42ff945cdf0d5205c6ca61d6861a0479492dc
2012-07-25 18:56:51 +09:00
Jean Chalard e9a86e2cdb Search bigrams for the lower case version of the word (A46)
...if there aren't any for the exact case version.

Bug: 6752830
Change-Id: I2737148b01ba04a64febe009ceb2ef53c265d224
2012-07-04 20:12:58 +09:00
satok 1bc038c5e4 Move correction state to stack memory
*Before
(0)  13.18 (0.01%)
(1)  93025.41 (62.06%)
(2)  10.75 (0.01%)
(3)  10.50 (0.01%)
(4)  117.50 (0.08%)
(5)  55678.98 (37.14%)
(6)  9.09 (0.01%)
(20) 883.84 (0.59%)
Total 149898.24 (sum of others 149749.25)

*After
(0)  17.41 (0.01%)
(1)  92673.41 (61.95%)
(2)  10.62 (0.01%)
(3)  10.37 (0.01%)
(4)  120.96 (0.08%)
(5)  55741.18 (37.26%)
(6)  11.01 (0.01%)
(20) 862.72 (0.58%)
Total 149595.52 (sum of others 149447.68)

Change-Id: Ia5a25a544fc388e4dab1e08d8f78d5117b249cf3
2012-06-14 15:57:28 -07:00
Jean Chalard e308459531 Compute the correct frequency for bigram prediction
Change-Id: I3196f48a0ca2ed5e94f430254d58e65d341398c8
2012-05-29 16:22:46 +09:00
Jean Chalard cb99376307 Fix a bug where the bigram freq would be underevaluated
The difference in score is not large, but it's still a bug

Change-Id: Ie22c2b6e1206e829c1c8af096469df05af14d47b
2012-05-29 16:04:07 +09:00
Jean Chalard 19ebd93646 Split a method to reconstruct freq from uni/bi freq
This has no impact at all on the logic.

Change-Id: I3788c8335cc193433ad9a7512b211a49bb2ffb02
2012-05-29 16:00:25 +09:00
Jean Chalard 402b057050 Fix two small possible bugs.
None of these had any real impact, but they were potential
liabilities for the future

Change-Id: I2de581f8b638e423d47a6d99b1a3c96af4c6150d
2012-05-29 15:56:30 +09:00
Jean Chalard 9416c81403 Return the bigram frequency if available.
This concludes the work on bug#6313806.
Don't submit it before the dictionaries are suitably amended.

Bug: 6313806
Change-Id: Icfea45bd52bb9d8cc68ba2266f80640e3942bb7f
2012-05-16 21:14:06 +09:00
Jean Chalard 49ba135fde Perform the actual bigram frequency lookup.
This still returns the unigram frequency, because the values stored
for bigrams in the dictionary are not ready to be returned in-place
instead of unigram values. Aside from this, the code is complete.

Bug: 6313806
Change-Id: If7bb7b644730782277f0f6663334c170b7fe13fb
2012-05-10 20:01:44 +09:00
Jean Chalard 8950ce6c44 Replace the bigram list position with the map and filter
Passing the position will not allow us a reasonable lookup
time. Replace this with a map and bloom filter for very fast
lookup.

Bug: 6313806
Change-Id: I3a61c0001cbc987c1c3c7b8df635d4590a370144
2012-05-07 17:15:21 +09:00
Jean Chalard 171d1809ff Add methods to inverse compute the probability.
For now the probability is just returned with the same
value it had, but this is some ground work that needs to be
done anyway.

Bug: 6313806
Change-Id: I9bb8b96b294109771208ade558c9ad56932d2f8e
2012-04-24 09:40:44 +09:00
Jean Chalard 522a04ea5b Pass words as int[] to the native code.
We need to get the bigrams during the call to getSuggestions for
bug#6313806. We already give an int[] to getSuggestions and we
wanted to get rid of char[]'s anyway because it doesn't work with
surrogate pairs, so here we go.

Bug: 6313806
Change-Id: I56ce99f1db6b3302cdf42f0527343bded837091e
2012-04-23 16:05:36 +09:00
Jean Chalard e81ac8baa0 Add a method to get the flags from a binary dictionary.
This method is not used yet

Change-Id: Ic15d3d423aff2c83c712bc0aa56571d30755e663
2012-04-06 18:34:22 +09:00
Jean Chalard 5b0761e6a9 Remove write-only stuff
Change-Id: I5ac8ab64c77a298502b3d063ea70db9b4da41716
2012-04-06 17:52:18 +09:00
Jean Chalard 9a933a742d Read shortcuts as strings in the dictionary.
This has no impact on performance.
Before:
(0)  9.61 (0.01%)
(1)  57514.58 (56.70%)
(2)  10.55 (0.01%)
(3)  10.79 (0.01%)
(4)  133.20 (0.13%)
(5)  43553.87 (42.94%)
(6)  10.03 (0.01%)
(20) 47.20 (0.05%)
Total 101431.47 (sum of others 101289.84)

After:
(0)  10.52 (0.01%)
(1)  56311.16 (56.66%)
(2)  13.40 (0.01%)
(3)  10.98 (0.01%)
(4)  136.72 (0.14%)
(5)  42707.92 (42.97%)
(6)  9.79 (0.01%)
(20) 51.35 (0.05%)
Total 99390.76 (sum of others 99251.84)

The difference is not significant with regard to measure imprecision

Change-Id: I2e4f1ef7a5e99082e67dd27f56cf4fc432bb48fa
2012-04-06 16:22:08 +09:00
Ken Wakasa 3ef3e24a12 Move the "src" directory as a preparation for Ib4a47342 and I66f6c5b9
Change-Id: I3ab65059f6e356530484bfd0bba26a634a4cba65
2012-03-30 09:53:51 +09:00