Commit Graph

650 Commits

Author SHA1 Message Date
yunfan
5241e30bdd - update transformer docs 2019-02-20 17:27:36 +08:00
Coet
767e7971e5
Merge pull request #134 from chenkaiyu1997/master
[new] Add ENAS (Efficient Neural Architecture Search)
2019-02-18 09:59:14 +08:00
chenkaiyu1997
efeac2c427 Add ENAS (Efficient Neural Architecture Search) 2019-02-15 17:31:56 +08:00
Coet
13faa2b410
Merge pull request #132 from FengZiYjun/v0.3.1
fastNLP V0.3.1
2019-02-06 10:40:53 +08:00
FengZiYjun
b66d7b8f51 update API introduction 2019-02-04 10:07:56 +08:00
FengZiYjun
d1b5adabc4 add codecov fix 2019-02-04 09:56:08 +08:00
FengZiYjun
0c5630bd16 Ready for V0.3.1
* 升级parser API和模型
* update docs: add new pages for tutorials
* upgrade CWS api download source
* add a new method for dataset field access
* add introduction for bert
* add more unit tests for api/processor
* remove unused test data. Add new test data.
2019-02-04 09:44:54 +08:00
FengZiYjun
986541139a 整理所有dataset loader,建立单元测试 2019-02-02 16:46:42 +08:00
FengZiYjun
bfaf09df8c add BERT model
* load pre-trained BERT weights from local binary
* add tests
2019-01-29 20:35:12 +08:00
FengZiYjun
887fc9281f update callbacks:
* rename callback methods. Use fastai's notation.
* add a new callback method - on_valid_begin
2019-01-25 21:43:24 +08:00
xuyige
e0d6a259ae skip training while n_epoch in trainer is not greater than 0 2019-01-23 17:09:28 +08:00
xuyige
d4b4ffa28b add testing tutorial 2019-01-23 14:56:25 +08:00
yunfan
9474ab4b34 remove device in batch 2019-01-21 22:28:31 +08:00
yunfan
c02980e006 Merge branch 'yyff' into dev 2019-01-21 14:55:53 +08:00
yunfan
a37de4344d add batch device 2019-01-21 14:55:32 +08:00
yunfan
e93c6f0053 Merge branch 'dev' of https://github.com/choosewhatulike/fastNLP-private into dev 2019-01-21 14:52:24 +08:00
yh
47ec69ea96 trainer根据syf的多进程batch进行修改 2019-01-21 14:44:31 +08:00
FengZiYjun
f3cb812554 将tesorboardX处理为callback, 从trainer移除tensorboardX相关代码 2019-01-20 16:37:58 +08:00
yunfan
03f49c8264 - batch with multiprocessing 2019-01-19 19:44:32 +08:00
FengZiYjun
b14dd58828 Update POS API 2019-01-19 18:48:57 +08:00
FengZiYjun
62ea4f7fed 添加LR finder,用第一个epoch找最佳lr,从第二个epoch开始训练 2019-01-19 18:40:43 +08:00
yunfan
a7f3701bdf - revert batch 2019-01-19 16:26:39 +08:00
yunfan
de856fb8eb update reproduction 2019-01-19 16:22:01 +08:00
yunfan
eb55856c78 - fix parser train 2019-01-19 16:07:10 +08:00
FengZiYjun
ab953b43ab * 重构POS API,改成接受word作为输入
* 添加两类Callback
* 完善Trainer对error的捕捉
2019-01-19 15:23:07 +08:00
yh_cc
d9ac334409 减少batch中不断创建多进程的开销 2019-01-18 23:33:19 +08:00
yh_cc
2e3ef52a7d 将batch增强为多进程batch 2019-01-18 23:02:15 +08:00
FengZiYjun
864c2238f8 添加FieldArray对list of np.array的支持 2019-01-17 22:42:40 +08:00
FengZiYjun
b93ca9bb30 * FieldArray添加对list of np.array的支持
* 添加测试:FieldArray的初始化
2019-01-17 15:39:13 +08:00
FengZiYjun
e4f997d52a refactor type system in FieldArray:
* 重构dtype的检测代码,在FieldArray的初始化和append两处,达到更好的代码复用
* 类型检测的责任完全落在FieldArray,DataSet与之配合
测试:
* 整理dtype相关的测试代码
* 给所有tutorial添加测试
其他:
* 完善一个完整的Conll dataset loader
* 升级POS tag model训练脚本
2019-01-17 12:25:37 +08:00
yh
73dd35dabf Merge branch 'dev' of github.com:choosewhatulike/fastNLP-private into dev 2019-01-15 22:23:35 +08:00
yh
3e33a23042 修改Padder的测试用例 2019-01-15 22:23:19 +08:00
yh
8091a734ee 1. 将pad的功能从FieldArray中剥离出来,使用Padder完成各种padding操作。
2. FieldArray默认使用AutoPadder, AutoPadder的行为与之前不使用padder是一致的的
3. 为了解决二维padding的问题,引入了EngChar2dPadder用于对character进行padding
4. 增加一份padding的tutorial。
2019-01-15 22:21:55 +08:00
FengZiYjun
d80d944e40 * 添加callbacks:EarlyStopCallback
* 将dataset.py中的assert改为raise error
* 给trainer添加try-except,捕捉EarlyStopError
* 优化trainer代码
* 给callbacks添加测试
2019-01-15 17:50:00 +08:00
yh
1f50b01ffa conflict solved 2019-01-15 15:16:20 +08:00
yh
6a0a1ed4ad train增加注释;attention增加注释;新增transformer分词 2019-01-15 14:58:43 +08:00
FengZiYjun
1fdaf236d2 Updates:
* 改名: chinese_word_segment ---> Chinese_word_segmentation
* 改名: pos_tag_model ---> POS_tagging
* 添加4个对Batch的测试
* 删除无用的chinese_word_segment/run.py
2019-01-15 14:56:01 +08:00
FengZiYjun
c4ba75d160 code optimization
* move used readers from reproduction to io/dataset_loader.py
(API shall not call anything from reproduction/)
2019-01-15 14:30:37 +08:00
ChenXin
a6dbbe9812 remove the gpu_id info when saving 2019-01-15 11:45:02 +08:00
yunfan
2e9e6c6c20 - fix trainer with validate_every > 0
- refine & fix Transformer Encoder
- refine & speed up biaffine parser
2019-01-14 19:13:52 +08:00
Coet
3fa95b6cfc
Update README.md
fix tutorial link
2019-01-14 09:31:35 +08:00
Coet
3219b9da33
Merge pull request #123 from FengZiYjun/v0.3.0
fastNLP V0.3.0
2019-01-14 09:07:31 +08:00
FengZiYjun
094a566155 add syntactic parsing API 2019-01-13 11:49:38 +08:00
FengZiYjun
8df5bce938 fastNLP V0.3 2019-01-12 19:15:20 +08:00
FengZiYjun
eca0c6f8c4 Prepare for release.
* pass all tests
* prepare CWS & POS API
* update tutorials
* add README.md in tutorials/ & api/
2019-01-12 19:00:06 +08:00
yunfan
62a7556a04 Merge remote-tracking branch 'private/dev' into dev
# Conflicts:
#	fastNLP/api/api.py
#	fastNLP/modules/encoder/variational_rnn.py
2019-01-12 11:26:32 +08:00
yunfan
ba28702e68 update Biaffine Parser, Variational RNN
add parser API
2019-01-12 11:22:09 +08:00
FengZiYjun
7eb02f1762 update code comments in CRF 2019-01-11 20:59:42 +08:00
FengZiYjun
751fe2768e Merge branch 'master' of https://github.com/fastnlp/fastNLP into current branch
# Conflicts solved:
#	fastNLP/io/dataset_loader.py
#	fastNLP/io/embed_loader.py
#	fastNLP/modules/aggregator/attention.py
#	fastNLP/modules/decoder/CRF.py
2019-01-11 20:44:26 +08:00
FengZiYjun
c0b67a2bc9 fix tests 2019-01-11 20:24:35 +08:00