Matthew Honnibal
|
9e960d24fc
|
Refactor conllu script, fix interface, generalize
|
2018-02-25 14:54:47 +01:00 |
|
Matthew Honnibal
|
551c93fe01
|
Shuffle data after each epoch. Improve script
|
2018-02-25 13:35:32 +01:00 |
|
Matthew Honnibal
|
bdb0174571
|
Update conllu training script
|
2018-02-25 13:12:39 +01:00 |
|
Matthew Honnibal
|
e09070eca7
|
Refactor conllu script
|
2018-02-25 12:50:29 +01:00 |
|
Matthew Honnibal
|
44e496a82e
|
Refactor conllu script
|
2018-02-25 12:48:22 +01:00 |
|
Matthew Honnibal
|
c388833ca6
|
Minibatch by number of tokens, support other vectors, refactor CoNLL printing
|
2018-02-25 10:38:06 +01:00 |
|
Matthew Honnibal
|
dd78ef066a
|
Unset data size limit in conll script
|
2018-02-24 18:14:57 +01:00 |
|
Matthew Honnibal
|
8adeea3746
|
Generalize conllu script. Now handling Chinese (maybe badly)
|
2018-02-24 16:04:27 +01:00 |
|
Matthew Honnibal
|
329b14c9e6
|
Clean up conllu script
|
2018-02-24 10:31:53 +01:00 |
|
Matthew Honnibal
|
5be092ee72
|
CONLLU scoring 80.9% UAS with no oracle segments
|
2018-02-23 23:49:17 +01:00 |
|
Matthew Honnibal
|
23236340f4
|
Update CoNLL script. Don't preset SBD. Set batch size to 8, avoid writing twice
|
2018-02-22 21:35:50 +01:00 |
|
Matthew Honnibal
|
a26e399f84
|
Update conllu script
|
2018-02-22 19:43:54 +01:00 |
|
Matthew Honnibal
|
001e2ec6d6
|
Refactor CoNLL training script
|
2018-02-22 16:00:34 +01:00 |
|
Matthew Honnibal
|
6a27a4f77c
|
Set accelerating batch size in CONLL train script
|
2018-02-21 21:02:41 +01:00 |
|
Matthew Honnibal
|
4dc0fc9954
|
Replace labels that didn't make freq cutoff
|
2018-02-21 15:59:22 +01:00 |
|
Matthew Honnibal
|
97164b1763
|
Fix conllu script
|
2018-02-21 14:46:54 +01:00 |
|
Matthew Honnibal
|
24fb2c246f
|
Add script to do conllu training
|
2018-02-21 13:53:59 +01:00 |
|