1033 Commits

Author SHA1 Message Date
Jeremy Howard
98dc6010de Merge pull request #448 from k0ala/master
fix typos
2018-05-08 11:39:59 -07:00
Jeremy Howard
963579f53b Merge pull request #452 from sbalk/patch-1
Fix import error for Inception v4
2018-05-08 11:35:30 -07:00
Jeremy Howard
44edf3acac Merge pull request #453 from sean-adler/fix-randomized-transform-tests
Fix randomized transform tests
2018-05-08 11:35:14 -07:00
Jeremy Howard
6c4d40fa54 Merge pull request #454 from sgugger/master
Adds use_swa and wd outside of the loss to the new training API
2018-05-08 11:34:30 -07:00
Sylvain Gugger
5f027d0729 Remove stock grads 2018-05-08 07:58:25 -04:00
Sylvain Gugger
cc28c537bf Adds use_swa and wd out of loss to the new training API
1. Adds the use_swa functionality inside the fit_opt_sched function
2.  Adds the option wd_loss (default True) to a the TrainingPhase object. If set to false, weight decay is computed outside of the loss function. Test on the training of a LM with the 1cycle policy shows similar performance so I think it's working properly. Needs to be tested with Adam on other datasets.
2018-05-07 10:58:23 -04:00
Sean Adler
943fa8c382 Apply padding to min and max coordinates of 'cutout()' holes 2018-05-06 18:10:15 -07:00
Sean Adler
e2e467f858 Use correct upper-bound for 'RandomScale' output shape 2018-05-06 18:09:08 -07:00
Stijn Balk
41d4797b82 Fix import error for Inception v4
The InceptionV4 class was imported instead of the inceptionv4 model.
2018-05-06 19:35:55 +02:00
k0ala
f19a76a46f fix typos 2018-05-05 09:08:36 +02:00
Jeremy Howard
5906daacb4 Merge pull request #445 from sudarshan85/jup_ext
Added conda-forge channel only for jup extensions
2018-05-04 14:05:18 -07:00
Jeremy Howard
54b0541eeb Merge pull request #446 from sivagnanamn/patch-1
Removed duplicate dependency
2018-05-04 14:04:44 -07:00
Sivagnanam Namasivayamurthy
42948c4636 Removed duplicate dependency 2018-05-04 22:38:12 +09:00
Sudarshan Srinivasan
7b2e3a0ab0 Added conda-forge channel only for jup extensions 2018-05-04 08:35:01 -04:00
Jeremy Howard
848405510b Merge pull request #437 from bearpelican/cifar10_dawn
Adding notebook for cifar10 dawn
2018-05-03 12:30:37 -07:00
Jeremy Howard
ddc8da7ebf Merge pull request #441 from NEIA20/adds-core-test-file
Adds test file for `core.py` and tests for `partition` fn
2018-05-03 12:21:21 -07:00
Nisreen Al-Basha
95c2ad2d6e improves core tests 2018-05-02 22:34:05 -04:00
Nisreen Al-Basha
90dac084c4 Cleans up 2018-05-02 21:59:17 -04:00
Nisreen Al-Basha
cd4db7c83a Test passing 2018-05-02 21:51:22 -04:00
Andrew Shaw
ddae62b158 Adding notebook for cifar10 dawn 2018-05-02 19:37:45 -04:00
Jeremy Howard
94036edada restrict pytorch <0.4 2018-05-02 15:29:01 -07:00
Jeremy Howard
d02cba7465 Merge branch 'master' of github.com:fastai/fastai 2018-05-02 14:49:43 -07:00
Jeremy Howard
dedf0236bc remove awscli 2018-05-02 14:49:38 -07:00
Jeremy Howard
e33a3e93b3 Merge pull request #416 from thedch/dataloader_doc
Dataloader doc
2018-05-02 14:46:50 -07:00
Jeremy Howard
41b512abe2 Merge pull request #428 from benediktbrandt/from_csv_asserts
Asserts for ImageClassifierData.from_csv
2018-05-02 14:42:57 -07:00
Jeremy Howard
fee6824e73 Merge pull request #430 from JanLauGe/master
modify LR_Finder.plot to use n_skip_end=0
2018-05-02 14:42:02 -07:00
Jeremy Howard
784d3f64cf Merge pull request #433 from PiotrCzapla/transforms-tests
Extract transforms tests & some fixes as separate PR from #312
2018-05-02 14:41:15 -07:00
Daniel Hunter
202860cdb9 Fixed argument formatting 2018-05-01 13:40:51 -07:00
Piotr Czapla
b5b8b09a08 Add few more tests showing how the current API works. 2018-05-01 19:21:14 +02:00
Piotr Czapla
91306381f8 Add docs to scale_to 2018-05-01 18:38:29 +02:00
Piotr Czapla
ce752f1a67 Deprecate unused parameter in to_bb fixes #432 2018-05-01 18:38:29 +02:00
Piotr Czapla
a32ce77755 Make RandomRotateZoom use do_transform
I don't see a reason why this wasn't using the `do_transform` as all
other classes and not using it breaks the new API, so I've fixed it
instead of adding a wrapper to new API
2018-05-01 18:28:21 +02:00
Piotr Czapla
e6c00d88bd Make RandomLighting respect tfm_y
Fixes #321
2018-05-01 18:28:21 +02:00
Piotr Czapla
f1576bb3d7 Add tests to new transforms.py 2018-05-01 18:26:01 +02:00
Piotr Czapla
92a2c55d66 Make sure that pytest does not discover test in courses
That folder contains files that pytest views as tests, but it can't import them.
Without this the exclusion the test discovery fails in Visual Studio Code
2018-05-01 18:07:29 +02:00
Laurens Geffert
25169975f4 modified LR_Finder.plot to use n_skip_end=0
modified learning rate finder method LR_Finder.plot so that `n_skip_end=0` plots all values. Previously, one had to use `n_skip_end=1` to get all values, while n_skip_end=0 would return an empty plot. The new behaviour makes `n_skip_end` consistent with the behaviour of `n_skip`
2018-05-01 15:44:01 +01:00
b1
1cc6b7d98b Added two asserts to make sure that ImageClassifierData.from_csv is correctly invoked 2018-05-01 09:00:36 -04:00
Jeremy Howard
750e810963 Merge branch 'master' of github.com:fastai/fastai 2018-04-30 18:13:04 -07:00
Jeremy Howard
64aea869ea lesson 14 2018-04-30 18:12:58 -07:00
Jeremy Howard
d96a806b92 xl 2018-04-30 18:12:27 -07:00
Jeremy Howard
e21b14c58d Merge pull request #420 from noklam/master
Change unfreeze to freeze last layer fine tuning
2018-04-30 13:28:33 -07:00
Jeremy Howard
03bf511176 Merge pull request #423 from narudesu/patch-1
Correct property typo (self.data.path)
2018-04-30 13:26:16 -07:00
Jeremy Howard
337509704a Merge pull request #425 from talhadar/patch-1
change using self.data_path to self.data.path
2018-04-30 13:25:32 -07:00
Tal Hadar
6492d01e1d change using self.data_path to self.data.path
using self.data_path in line 37 causes an error: AttributeError: ‘ConvLearner’ object has no attribute ‘data_path’
2018-04-30 19:38:52 +03:00
Matěj Šmíd
09b8b92c59 Correct property typo (self.data.path)
I got an error when I ran the tutorial code of deep learning 1. It said that self.data_path was not defined, so I looked into learner.py and saw that on the line before this line, there was self.data.path used instead of self.data_path, so I believe that should be corrected.
2018-04-30 18:27:48 +02:00
GFK\noklam.chan
9f3666818c Correct unfreeze to freeze last layer 2018-04-30 17:33:04 +08:00
Daniel Hunter
81383af289 Add the word method to methods 2018-04-29 19:40:42 -07:00
Daniel Hunter
b3796fb0cd Add dataloader doc 2018-04-29 19:36:08 -07:00
Jeremy Howard
7d0a033cec Merge pull request #411 from mcskinner/sampler-comments
add inline comments to SortishSampler biggest batch logic
2018-04-29 15:24:28 -07:00
Jeremy Howard
d56de94ae3 Merge pull request #410 from mcskinner/whitespace-fix
fix whitespace issues in dataset.py
2018-04-29 15:23:51 -07:00