Commit Graph

162 Commits

Author SHA1 Message Date
Rob Speer
a893823d6e un-flake wordfreq_builder.tokenizers, and edit docstrings 2015-08-26 13:03:23 -04:00
Rob Speer
5a1fc00aaa Strip apostrophes from edges of tokens
The issue here is that if you had French text with an apostrophe,
such as "d'un", it would split it into "d'" and "un", but if "d'"
were re-tokenized it would come out as "d". Stripping apostrophes
makes the process more idempotent.
2015-08-25 12:41:48 -04:00
Rob Speer
de73888a76 use better regexes in wordfreq_builder tokenizer 2015-08-24 19:05:46 -04:00
Rob Speer
140ca6c050 remove Hangul fillers that confuse cld2 2015-08-24 17:11:18 -04:00
Andrew Lin
6d40912ef9 Stylistic cleanups to word_counts.py. 2015-07-31 19:26:18 -04:00
Andrew Lin
53621c34df Remove redundant reference to wikipedia in builder README. 2015-07-31 19:12:59 -04:00
Rob Speer
e9f9c94e36 Don't use the file-reading cutoff when writing centibels 2015-07-28 18:45:26 -04:00
Rob Speer
c5708b24e4 put back the freqs_to_cBpack cutoff; prepare for 1.0 2015-07-28 18:01:12 -04:00
Rob Speer
32102ba3c2 Merge pull request #19 from LuminosoInsight/code-review-fixes-2015-07-17
Code review fixes 2015 07 17
2015-07-22 15:09:00 -04:00
Joshua Chin
93cd902899 updated read_freqs docs 2015-07-22 10:06:16 -04:00
Joshua Chin
4fe9d110e1 fixed style 2015-07-22 10:05:11 -04:00
Joshua Chin
6453d864c4 reordered command line args 2015-07-22 10:04:14 -04:00
Joshua Chin
8081145922 bugfix 2015-07-21 10:12:56 -04:00
Joshua Chin
c5f82ecac1 fixed rules.ninja 2015-07-20 17:20:29 -04:00
Joshua Chin
643571c69c fixed build bug 2015-07-20 16:51:25 -04:00
Joshua Chin
173278fdd3 ensure removal of tatweels (hopefully) 2015-07-20 16:48:36 -04:00
Joshua Chin
298d3c1d24 unhoisted if statement 2015-07-20 11:10:41 -04:00
Joshua Chin
accb7e398c ninja.py is now pep8 compliant 2015-07-20 11:06:58 -04:00
Joshua Chin
221acf7921 fixed build 2015-07-17 17:44:01 -04:00
Rob Speer
2d1020daac mention the Wikipedia data, and credit Hermit Dave 2015-07-17 17:09:36 -04:00
Joshua Chin
f31f9a1bcd fixed tokenize_twitter 2015-07-17 16:37:47 -04:00
Joshua Chin
a44927e98e added cld2 tokenizer comments 2015-07-17 16:03:33 -04:00
Joshua Chin
11a1c51321 fix arabic tokens 2015-07-17 15:52:12 -04:00
Joshua Chin
c75c735d8d fixed syntax 2015-07-17 15:43:24 -04:00
Joshua Chin
303bd88ba2 renamed tokenize file to tokenize twitter 2015-07-17 15:27:26 -04:00
Joshua Chin
d6519cf736 created last_tab flag 2015-07-17 15:19:09 -04:00
Joshua Chin
620becb7e8 removed uncessary if statement 2015-07-17 15:14:06 -04:00
Joshua Chin
d988b1b42e generated freq dict in place 2015-07-17 15:13:25 -04:00
Joshua Chin
e37c689031 corrected docstring 2015-07-17 15:12:23 -04:00
Joshua Chin
002351bace removed unnecessary strip 2015-07-17 15:11:28 -04:00
Joshua Chin
7fc23666a9 moved last_tab to tokenize_twitter 2015-07-17 15:10:17 -04:00
Joshua Chin
528285a982 removed unused function 2015-07-17 15:03:14 -04:00
Joshua Chin
59d3c72758 fixed spacing 2015-07-17 15:02:34 -04:00
Joshua Chin
10028be212 removed unnecessary format 2015-07-17 15:01:25 -04:00
Joshua Chin
3b368b66dd cleaned up BAD_CHAR_RANGE 2015-07-17 15:00:59 -04:00
Joshua Chin
c2d1cdcb31 moved test tokenizers 2015-07-17 14:58:58 -04:00
Joshua Chin
5d26c9f57f added docstring and moved to scripts 2015-07-17 14:56:18 -04:00
Joshua Chin
bdc791af8f style changes 2015-07-17 14:54:32 -04:00
Joshua Chin
4d5ec57144 removed bad comment 2015-07-17 14:54:09 -04:00
Joshua Chin
39f01b0485 removed unused scripts 2015-07-17 14:53:18 -04:00
Joshua Chin
98a7a8093b removed mkdir -p for many cases 2015-07-17 14:45:22 -04:00
Joshua Chin
449a656edd removed TOKENIZE_TWITTER 2015-07-17 14:43:14 -04:00
Joshua Chin
00e18b7d4b removed TOKENIZE_TWITTER option 2015-07-17 14:40:49 -04:00
Joshua Chin
772c0cddd1 more README fixes 2015-07-17 14:40:33 -04:00
Joshua Chin
0a085132f4 fixed README 2015-07-17 14:35:43 -04:00
Rob Speer
8633e8c2a9 update the wordfreq_builder README 2015-07-13 11:58:48 -04:00
Rob Speer
41dba74da2 add docstrings and remove some brackets 2015-07-07 18:22:51 -04:00
Joshua Chin
b0f759d322 Removes mention of Rosette from README 2015-07-07 10:32:16 -04:00
Rob Speer
10c04d116f add 'twitter' as a final build, and a new build dir
The `data/dist` directory is now a convenient place to find the final
built files that can be copied into wordfreq.
2015-07-01 17:45:39 -04:00
Rob Speer
37375383e8 cope with occasional Unicode errors in the input 2015-06-30 17:05:40 -04:00
Rob Speer
4771c12814 remove wiki2tokens and tokenize_wikipedia
These components are no longer necessary. Wikipedia output can and
should be tokenized with the standard tokenizer, instead of the
almost-equivalent one in the Nim code.
2015-06-30 15:28:01 -04:00
Rob Speer
9a2855394d fix comment and whitespace involving tokenize_twitter 2015-06-30 15:18:37 -04:00
Rob Speer
f305679caf Switch to a centibel scale, add a header to the data 2015-06-22 17:38:13 -04:00
Rob Speer
d16683f2b9 Merge pull request #2 from LuminosoInsight/review-refactor
Adds a number of bugfixes and improvements to wordfreq_builder
2015-06-19 15:29:52 -04:00
Rob Speer
5bc1f0c097 restore missing Russian OpenSubtitles data 2015-06-19 12:36:08 -04:00
Joshua Chin
3746af1350 updated freqs_to_dBpack docstring 2015-06-18 10:32:53 -04:00
Joshua Chin
59ce14cdd0 revised read_freqs docstring 2015-06-18 10:28:22 -04:00
Joshua Chin
04bf6aadcc updated monolingual_tokenize_file docstring, and removed unused argument 2015-06-18 10:20:54 -04:00
Joshua Chin
91dd73a2b5 tokenize_file should ignore lines with unknown languages 2015-06-18 10:18:57 -04:00
Joshua Chin
ffc01c75a0 Fixed CLD2_BAD_CHAR regex 2015-06-18 10:18:00 -04:00
Joshua Chin
8277de2c7f changed tokenize_file: cld2 return 'un' instead of None if it cannot recognize the language 2015-06-17 14:19:28 -04:00
Joshua Chin
b24f31d30a tokenize_file: don't join tokens if language is None 2015-06-17 14:18:18 -04:00
Joshua Chin
99d97956e6 automatically closes input file in tokenize_file 2015-06-17 11:42:34 -04:00
Joshua Chin
e50c0c6917 updated test to check number parsing 2015-06-17 11:30:25 -04:00
Joshua Chin
c71e93611b fixed build process 2015-06-17 11:25:07 -04:00
Joshua Chin
8317ea6d51 updated directory of twitter output 2015-06-16 17:32:58 -04:00
Joshua Chin
da93bc89c2 removed intermediate twitter file rules 2015-06-16 17:28:09 -04:00
Joshua Chin
87f08780c8 improved tokenize_file and updated docstring 2015-06-16 17:27:27 -04:00
Joshua Chin
bea8963a79 renamed pretokenize_twitter to tokenize twitter, and deleted format_twitter 2015-06-16 17:26:52 -04:00
Joshua Chin
aeedb408b7 fixed bugs and removed unused code 2015-06-16 17:25:06 -04:00
Joshua Chin
64644d8ede changed tokenizer to only strip t.co urls 2015-06-16 16:11:31 -04:00
Joshua Chin
b649d45e61 Added codepoints U+10FFFE and U+10FFFF to CLD2_BAD_CHAR_RANGE 2015-06-16 16:03:58 -04:00
Joshua Chin
a200a0a689 added tests for the tokenizer and language recognizer 2015-06-16 16:00:14 -04:00
Joshua Chin
1cf7e3d2b9 added pycld2 dependency 2015-06-16 15:06:22 -04:00
Joshua Chin
297d981e20 Replaced Rosette with cld2 language recognizer and wordfreq tokenizer 2015-06-16 14:45:49 -04:00
Rob Speer
b78d8ca3ee ninja2dot: make a graph of the build process 2015-06-15 13:14:32 -04:00
Rob Speer
56d447a825 Reorganize and document some functions 2015-06-15 12:40:31 -04:00
Rob Speer
3d28491f4d okay, apparently you can't mix code blocks and bullets 2015-06-01 11:39:42 -04:00
Rob Speer
d202474763 is this indented enough for you, markdown 2015-06-01 11:38:10 -04:00
Rob Speer
9927a8c414 add a README 2015-06-01 11:37:19 -04:00
Rob Speer
cbe3513e08 Tokenize Japanese consistently with MeCab 2015-05-27 17:44:58 -04:00
Rob Speer
536c15fbdb give mecab a larger buffer 2015-05-26 19:34:46 -04:00
Rob Speer
5de81c7111 fix build rules for Japanese Wikipedia 2015-05-26 18:08:57 -04:00
Rob Speer
3d5b3d47e8 fix version in config.py 2015-05-26 18:08:46 -04:00
Rob Speer
ffd352f148 correct a Leeds bug; add some comments to rules.ninja 2015-05-26 18:08:04 -04:00
Rob Speer
50ff85ce19 add Google Books data for English 2015-05-11 18:44:28 -04:00
Rob Speer
c707b32345 move some functions to the wordfreq package 2015-05-11 17:02:52 -04:00
Rob Speer
d0d777ed91 use a more general-purpose tokenizer, not 'retokenize' 2015-05-08 12:40:14 -04:00
Rob Speer
35128a94ca build.ninja knows about its own dependencies 2015-05-08 12:40:06 -04:00
Rob Speer
d6cc90792f Makefile should only be needed for bootstrapping Ninja 2015-05-08 12:39:31 -04:00
Rob Speer
2f14417bcf limit final builds to languages with >= 2 sources 2015-05-07 23:59:04 -04:00
Rob Speer
1b7a2b9d0b fix dependency 2015-05-07 23:55:57 -04:00
Rob Speer
abb0e059c8 a reasonably complete build process 2015-05-07 19:38:33 -04:00
Rob Speer
02d8b32119 process leeds and opensubtitles 2015-05-07 17:07:33 -04:00
Rob Speer
7e238cf547 abstract how we define build rules a bit 2015-05-07 16:59:28 -04:00
Rob Speer
d2f9c60776 WIP on more build steps 2015-05-07 16:49:53 -04:00
Rob Speer
16928ed182 add rules to count wikipedia tokens 2015-05-05 15:21:24 -04:00
Rob Speer
bd579e2319 fix the 'count' ninja rule 2015-05-05 14:06:13 -04:00
Rob Speer
5787b6bb73 add and adjust some build steps
- more build steps for Wikipedia
- rename 'tokenize_twitter' to 'pretokenize_twitter' to indicate that
  the results are preliminary
2015-05-05 13:59:21 -04:00
Rob Speer
61b9440e3d add wiki-parsing process 2015-05-04 13:25:01 -04:00