7a03662 | Luke Yeager | 04 November 2015, 00:36:51 UTC | Mark v3.0.0-rc.1 | 04 November 2015, 00:36:51 UTC |
c4cc194 | Luke Yeager | 04 November 2015, 00:14:43 UTC | Merge pull request #329 from jmancewicz/python-layer Add widget to copy python file from client or server to job dir | 04 November 2015, 00:14:43 UTC |
2b44d5d | Joe Mancewicz | 23 September 2015, 02:03:10 UTC | Add widget to copy python file from client or server to job dir | 03 November 2015, 20:53:13 UTC |
b07b425 | Luke Yeager | 03 November 2015, 17:52:06 UTC | Merge pull request #392 from lukeyeager/use-different-port Use port 34448 (DIGIT) for digits-server, not 8080 | 03 November 2015, 17:52:06 UTC |
6dfa82e | Luke Yeager | 03 November 2015, 17:51:50 UTC | Merge pull request #397 from lukeyeager/revert-pr394 Revert "Reduce Torch7 install verbosity" | 03 November 2015, 17:51:50 UTC |
c9e2a0f | Luke Yeager | 03 November 2015, 17:48:12 UTC | Revert "Reduce Torch7 install verbosity" This reverts commit a25e13acd0c24178a18012a706505f57eaeb894d. | 03 November 2015, 17:48:12 UTC |
17e0cd3 | Luke Yeager | 02 November 2015, 20:26:00 UTC | Merge pull request #395 from gheinrich/dev/no-crop-in-python Don't crop image before feeding to DL framework | 02 November 2015, 20:26:00 UTC |
1437388 | Greg Heinrich | 02 November 2015, 10:24:39 UTC | Don't crop image before feeding to DL framework When using 'crop in form' method, images are cropped in Python code before 'classify one'. This is slightly inconsistent as this operation is not performed during training or in the context of 'classify many'. Besides, this is problematic for Torch (at least) since the mean image is not cropped and this makes it difficult to perform mean subtraction during 'classify one' as the input and mean image have different sizes. Caffe uses the 'crop_size' option in the data layer transformations. Similarly, for Torch the crop command-line flag may be used to crop the image within Torch. Either way, it is not necessary to crop the image in Python code. | 02 November 2015, 19:53:13 UTC |
98b7298 | Luke Yeager | 02 November 2015, 16:55:59 UTC | Merge pull request #394 from gheinrich/dev/torch-install-verbosity Reduce Torch7 install verbosity | 02 November 2015, 16:55:59 UTC |
a25e13a | Greg Heinrich | 01 November 2015, 20:52:47 UTC | Reduce Torch7 install verbosity On DIGITS ToT, Travis logs are close to 4MB in size, which is the hard limit for log size on Travis. Longer logs would yield an error. This change causes the Torch7 install log to only be printed on an install error. This reduces the size of a typical log down to 250kB and 4000 lines, meaning that the full log can be displayed in real time since it is less than 10,000 lines. | 01 November 2015, 21:20:46 UTC |
2843680 | Luke Yeager | 30 October 2015, 23:50:11 UTC | Use port 34448 (DIGIT) for digits-server, not 8080 Port 8080 is a fairly common debugging port. Switch to using 34448 instead (that's "DIGIT" on an old-school phone). The user shouldn't need to be aware of the port at all, since digits-server is intended to run behind an nginx reverse proxy (see nginx.site). | 30 October 2015, 23:50:11 UTC |
a483351 | Luke Yeager | 30 October 2015, 23:26:31 UTC | Merge pull request #390 from gheinrich/dev/multi-threaded-test Torch multi-threaded data loader | 30 October 2015, 23:26:31 UTC |
038840d | Greg Heinrich | 09 October 2015, 16:19:45 UTC | Torch multi-threaded data loader Using MNIST (45k training samples, 15k validation samples) and LeNet (20 epochs): original: 118s new: 91s Using CIFAR10 (images upscaled to 256x256): Alexnet (2 epochs) original: 948s Alexnet (2 epochs) new: 869s GoogLenet (2 epochs) original: 5280s GoogLenet (2 epochs) new: 5100s Note: occasionally a reader thread might fail to initialize in which case it will be disabled and a message like the one below will be printed: ``` [ERROR] thr-id=1 will be disabled - failed to initialize DB - #keys=14959 #records=14998 ``` | 30 October 2015, 22:34:12 UTC |
dd9927b | Joe Mancewicz | 30 October 2015, 01:04:24 UTC | Merge pull request #321 from jmancewicz/mean Make it optional to subtract mean pixel or mean image | 30 October 2015, 01:04:24 UTC |
a0286d3 | Joe Mancewicz | 06 October 2015, 01:37:33 UTC | Make it optional to subtract mean pixel or mean image #321 | 29 October 2015, 22:06:05 UTC |
0e60adf | Luke Yeager | 29 October 2015, 19:54:41 UTC | Merge pull request #387 from gheinrich/dev/openmp-disable Disable OPENMP on Travis | 29 October 2015, 19:54:41 UTC |
8371c51 | Greg Heinrich | 28 October 2015, 08:05:10 UTC | Disable OpenMP on Travis Avoid OpenBLAS+OpenMP instabilities. Reference: https://github.com/torch/ezinstall/commit/251fdab2d2210c3f76c510d82ba36ff8c5c55cca | 28 October 2015, 21:34:14 UTC |
5291984 | Luke Yeager | 27 October 2015, 17:26:05 UTC | Merge pull request #383 from lukeyeager/test-print-caffe-output Print raw caffe output for failed tests | 27 October 2015, 17:26:05 UTC |
6ba2b99 | Luke Yeager | 27 October 2015, 16:53:29 UTC | Print raw caffe output for failed tests - fix #377 Don't print the traceback - print the raw output. | 27 October 2015, 16:53:29 UTC |
f4a0d00 | Joe Mancewicz | 23 October 2015, 16:53:20 UTC | Merge pull request #334 from jmancewicz/clone Add ability to clone previous jobs | 23 October 2015, 16:53:20 UTC |
bf61b65 | Joe Mancewicz | 09 October 2015, 21:25:47 UTC | Added Clone Button to Job pages. | 23 October 2015, 16:37:33 UTC |
05d2e8c | Luke Yeager | 23 October 2015, 02:43:41 UTC | Merge pull request #380 from jmancewicz/remove-ago remove ago from running jobs | 23 October 2015, 02:43:41 UTC |
6282a4c | Joe Mancewicz | 22 October 2015, 22:13:02 UTC | remove ago from running jobs | 22 October 2015, 23:21:22 UTC |
87bd19e | Joe Mancewicz | 22 October 2015, 18:46:59 UTC | Merge pull request #240 from jmancewicz/job_management Job management | 22 October 2015, 18:46:59 UTC |
2bb22ee | Joe Mancewicz | 25 August 2015, 01:35:46 UTC | Adding Job Management UI | 22 October 2015, 18:02:59 UTC |
e525c6e | Luke Yeager | 21 October 2015, 16:34:50 UTC | Merge pull request #378 from gheinrich/dev/bug-346 Use fixed seed in model tests | 21 October 2015, 16:34:50 UTC |
3be421a | Greg Heinrich | 20 October 2015, 10:22:57 UTC | Use fixed seed in model tests Suggested by @lukeyeager to reduce the amount of non determinism in those tests. Tested with: ./digits-test -v --with-coverage --cover-package=digits -s 100 iterations of the full test suite.(83400 tests in total) 4 failures: digits.model.images.classification.test_views.TestCaffeLeNet.test_classify_one ... FAIL digits.model.images.classification.test_views.TestCaffeLeNet.test_classify_one_json ... FAIL FAIL: digits.model.images.classification.test_views.TestTorchCreatedHdf5Shuffle.test_classify_one FAIL: digits.model.images.classification.test_views.TestTorchCreatedHdf5Shuffle.test_classify_one_json | 21 October 2015, 07:49:02 UTC |
4bc6680 | Luke Yeager | 19 October 2015, 22:12:46 UTC | Merge pull request #377 from jmancewicz/print-traceback print the traceback to stdout on error | 19 October 2015, 22:12:46 UTC |
a2a1f4a | Joe Mancewicz | 19 October 2015, 21:54:36 UTC | print the traceback to stdout on error | 19 October 2015, 21:54:36 UTC |
8a7a1e0 | Luke Yeager | 19 October 2015, 18:12:46 UTC | Merge pull request #374 from lukeyeager/fix-status-test Fixup test_status.py | 19 October 2015, 18:12:46 UTC |
a7549bd | Luke Yeager | 19 October 2015, 17:05:46 UTC | Merge pull request #376 from gheinrich/dev/torch-diverge Fix parsing of infinite loss in Torch | 19 October 2015, 17:05:46 UTC |
58e07e8 | Greg Heinrich | 18 October 2015, 08:59:03 UTC | Fix parsing of infinite loss in Torch Lua uses 'inf' to represent infinite numbers. Additionally Torch may use 'nan' in its representation of infinite values in Tensors. We must catch both cases in the parsing of training and validation losses. Add test to verify functionality. | 19 October 2015, 09:18:57 UTC |
9b2ebb6 | Luke Yeager | 16 October 2015, 21:14:49 UTC | Fixup test_status.py Stop creating "tmp.p" files in the main DIGITS directory | 16 October 2015, 21:16:35 UTC |
831d929 | Luke Yeager | 16 October 2015, 16:37:55 UTC | Merge pull request #372 from lukeyeager/increase-timeouts Increase test timeouts | 16 October 2015, 16:37:55 UTC |
fcbc6ed | Luke Yeager | 16 October 2015, 16:34:12 UTC | Increase test timeouts | 16 October 2015, 16:34:12 UTC |
9230f5c | Luke Yeager | 16 October 2015, 16:19:52 UTC | Merge pull request #371 from gheinrich/dev/torch-batch-size Torch: support batch size specification in model file | 16 October 2015, 16:19:52 UTC |
2c006bb | Luke Yeager | 16 October 2015, 16:14:49 UTC | Merge pull request #370 from gheinrich/dev/torch-hdf5-bug Fix Torch HDF5 loading from multiple databases | 16 October 2015, 16:14:49 UTC |
58f4354 | Greg Heinrich | 06 October 2015, 21:05:44 UTC | Torch: support batch size specification in model file The default batch size in the DIGITS Lua wrappers is 16. This change allows specification of different batch sizes from model description for the training and validation phases. Also align batch sizes of Torch standard networks with Caffe standard networks. Using MNIST (45k training samples, 15k validation samples) and LeNet (20 epochs): default batch size (16): 190s default batch size from model (training: 64, validation: 100): 118s Using CIFAR10 (images upscaled to 256x256): Alexnet (2 epochs) default batch size (16): 1230s Alexnet (2 epochs) batch size from model (100): 948s GoogLenet (2 epochs) default batch size (16): 5340s GoogLenet (2 epochs) batch size from model (24): 5280s | 16 October 2015, 11:38:49 UTC |
de567d7 | Greg Heinrich | 15 October 2015, 15:37:01 UTC | Fix Torch HDF5 loading from multiple databases Variable name conflict prevents from loading HDF5 datasets that spread across several databases. Regression introduced with 13d28b | 16 October 2015, 11:25:57 UTC |
58944c4 | Luke Yeager | 15 October 2015, 18:06:45 UTC | Merge pull request #367 from lukeyeager/fix-torch-tempfiles Fix some of the open file issues with Torch | 15 October 2015, 18:06:45 UTC |
c473812 | Greg Heinrich | 15 October 2015, 11:27:03 UTC | Fix more OS handle leaks | 15 October 2015, 11:27:03 UTC |
efac4ee | Luke Yeager | 15 October 2015, 00:29:20 UTC | Fix some of the open file issues with Torch When running the image classification model tests like this: ./digits-test digits/model/images/classification/test_views.py the number of open files for the test suite grows to > 900. You can verify this with: while :; do lsof -p $PID | grep tmp | wc -l; sleep 1; done where PID is the process ID of the test suite. This patch brings the number of open files down from > 900 to ~90. There's more work to be done to fix this completely, but this takes a big step in the right direction. NOTE: I can't actually run the whole testsuite or I get a bunch of these OSErrors: OSError: [Errno 24] Too many open files | 15 October 2015, 00:30:54 UTC |
6ee6233 | Luke Yeager | 14 October 2015, 20:46:07 UTC | Merge pull request #366 from lukeyeager/fix-inference-error-display Display InferenceErrors like all other exceptions | 14 October 2015, 20:46:07 UTC |
d809d6b | Luke Yeager | 14 October 2015, 20:14:52 UTC | Display InferenceErrors like all other exceptions | 14 October 2015, 20:14:52 UTC |
e47d871 | Luke Yeager | 14 October 2015, 16:59:56 UTC | Merge pull request #362 from lukeyeager/fix-classify-many-ground-truth Ignore invalid ground truth for classify_many | 14 October 2015, 16:59:56 UTC |
9f9b9fe | Luke Yeager | 14 October 2015, 16:59:28 UTC | Merge pull request #364 from gheinrich/dev/bug-363 Skip Torch generic model tests if Torch not installed | 14 October 2015, 16:59:28 UTC |
9a95161 | Greg Heinrich | 14 October 2015, 08:13:30 UTC | Skip Torch generic model tests if Torch not installed closes #363 | 14 October 2015, 08:13:30 UTC |
765ade9 | Luke Yeager | 13 October 2015, 23:09:05 UTC | Ignore invalid ground truth for classify_many If you upload a file like: /path/to/image1.jpg 100 To a model with only 10 classes, DIGITS would throw an error. This patch sets the ground truth for that image to None instead. | 14 October 2015, 01:35:07 UTC |
893663e | Luke Yeager | 13 October 2015, 21:58:03 UTC | Merge pull request #361 from lukeyeager/fix-infer-many Fix infer_many for Caffe | 13 October 2015, 21:58:03 UTC |
9a5acc0 | Luke Yeager | 13 October 2015, 21:20:25 UTC | Fix infer_many for Caffe Previously, `scores = output` performed a shallow copy that was overwritten when the next output was calculated. Now, we're using np.copy to do a deep copy and avoid the error. | 13 October 2015, 21:20:25 UTC |
bb2b76f | Luke Yeager | 13 October 2015, 17:07:41 UTC | Merge pull request #345 from gheinrich/dev/torch_generic Torch generic inference | 13 October 2015, 17:07:41 UTC |
95359d8 | Luke Yeager | 13 October 2015, 17:07:06 UTC | Merge pull request #359 from lukeyeager/graphviz-optional Skip tests which fail because GraphViz is missing | 13 October 2015, 17:07:06 UTC |
01d0d14 | Luke Yeager | 13 October 2015, 00:47:56 UTC | Skip tests which fail because GraphViz is missing GraphViz is listed as optional in the README | 13 October 2015, 00:47:56 UTC |
afbefe3 | Luke Yeager | 12 October 2015, 22:21:18 UTC | Merge pull request #351 from lukeyeager/fix-bash-scripts Fix checks in bash script to throw useful errors | 12 October 2015, 22:21:18 UTC |
f17995f | Luke Yeager | 12 October 2015, 17:50:15 UTC | Merge pull request #358 from lukeyeager/remove-vgg Remove VGG network stub | 12 October 2015, 17:50:15 UTC |
88c9ee6 | Luke Yeager | 12 October 2015, 17:48:30 UTC | Remove VGG network stub | 12 October 2015, 17:48:30 UTC |
a056079 | Luke Yeager | 12 October 2015, 17:41:45 UTC | Merge pull request #357 from lukeyeager/increase-timeouts Increase timeouts | 12 October 2015, 17:41:45 UTC |
fc250fd | Greg Heinrich | 11 October 2015, 16:08:53 UTC | Increase timeouts | 12 October 2015, 17:39:26 UTC |
683c559 | Luke Yeager | 12 October 2015, 16:53:55 UTC | Merge pull request #352 from lukeyeager/remove-walkthrough Remove long-forgotten "digits-walkthrough" | 12 October 2015, 16:53:55 UTC |
f63e050 | Greg Heinrich | 11 October 2015, 16:08:53 UTC | Increase timeouts | 11 October 2015, 18:12:27 UTC |
13d28b4 | Greg Heinrich | 08 October 2015, 19:42:51 UTC | Torch unsupervised learning | 11 October 2015, 18:10:36 UTC |
22f1cf9 | Luke Yeager | 10 October 2015, 00:45:09 UTC | Merge pull request #350 from lukeyeager/path-manipulation Stop tools from importing global DIGITS install | 10 October 2015, 00:45:09 UTC |
bbb501f | Luke Yeager | 10 October 2015, 00:30:51 UTC | Remove long-forgotten "digits-walkthrough" | 10 October 2015, 00:30:51 UTC |
7d34ab5 | Luke Yeager | 10 October 2015, 00:27:16 UTC | Fix checks in bash script to throw useful errors Previously, if you didn't have nosetests installed, ./digits-test would fail like this: $ ./digits-test Unknown option: -- usage: /home/lyeager/digits/venv/bin/python [option] ... [-c cmd | -m mod | file | -] [arg] ... Try `python -h' for more information. Now, it would fail like this: $ ./digits-test ERROR: "nosetests" executable not found! | 10 October 2015, 00:28:27 UTC |
5a34077 | Greg Heinrich | 28 September 2015, 20:40:30 UTC | Torch generic inference | 08 October 2015, 20:00:03 UTC |
3e47f0f | Luke Yeager | 08 October 2015, 19:04:08 UTC | Stop tools from importing global DIGITS install If you have the digits package installed globally (on your PYTHONPATH) and you try to run a local DIGITS instance, the tools pick up the global installation instead of the local one when you start them in a new process. This patch fixes the issue. NOTE: in the future, this should be handled by moving the tools scripts inside the digits package. This is a short-term solution. | 08 October 2015, 19:08:21 UTC |
2daddc0 | Luke Yeager | 08 October 2015, 18:10:54 UTC | Merge pull request #348 from gheinrich/dev/torch-classification-tests More robust Torch image classification test | 08 October 2015, 18:10:54 UTC |
470ac0f | Greg Heinrich | 07 October 2015, 07:55:58 UTC | More robust Torch image classification test - using a color model of LeNet to more accurate discrimination of color images. - add more combinations of training parameters (DB backend and shuffle). - test correct classification of all categories | 07 October 2015, 07:55:58 UTC |
3a0ec3b | Luke Yeager | 06 October 2015, 18:20:04 UTC | Merge pull request #341 from lukeyeager/new-multi-gpu-flag Use new multi-GPU flag for NVcaffe v0.14 | 06 October 2015, 18:20:04 UTC |
55a1a00 | Luke Yeager | 06 October 2015, 17:31:18 UTC | Merge pull request #344 from gheinrich/dev/lenet-cudnn Torch7/LeNet: use cudnn when available | 06 October 2015, 17:31:18 UTC |
cbf36d1 | Luke Yeager | 02 October 2015, 22:13:29 UTC | Use new multi-GPU flag for v0.14 Upstream caffe decided to use the "--gpu" flag instead of the original "--gpus" flag which was merged into NVcaffe. | 06 October 2015, 17:19:32 UTC |
435b462 | Luke Yeager | 06 October 2015, 17:15:48 UTC | Merge pull request #331 from Deepomatic/datasetviz Dataset visualization | 06 October 2015, 17:15:48 UTC |
f2639b5 | groar | 04 October 2015, 19:26:20 UTC | Dataset exploration regenerated doc | 05 October 2015, 21:25:07 UTC |
93dfa0e | Greg Heinrich | 04 October 2015, 13:34:07 UTC | Torch7/LeNet: use cudnn when available On MNIST (1x28x28, 45k training images, 15k validation images), with batch size of 64. Training time for 20 epochs: Original: 8m26s With CuDNN: 3m7s | 04 October 2015, 13:34:07 UTC |
95f6a38 | Luke Yeager | 03 October 2015, 00:00:22 UTC | Merge pull request #340 from lukeyeager/fix-bs4-warning Suppress annoying BeautifulSoup warning | 03 October 2015, 00:00:22 UTC |
3ed5288 | Luke Yeager | 02 October 2015, 23:36:23 UTC | Merge pull request #342 from lukeyeager/reorder-apt-get-packages Reorder apt-get packages into logical progression | 02 October 2015, 23:36:23 UTC |
e8000e5 | Luke Yeager | 02 October 2015, 22:40:04 UTC | Reorder apt-get packages into logical progression | 02 October 2015, 22:40:04 UTC |
d0e260b | Luke Yeager | 02 October 2015, 22:12:24 UTC | Merge pull request #338 from lukeyeager/parse-version-v2 Fix warning with setuptools >= 8.0 | 02 October 2015, 22:12:24 UTC |
48d44b1 | Luke Yeager | 02 October 2015, 21:53:39 UTC | Suppress annoying BeautifulSoup warning There were a lot of warnings like this in the build logs: /home/travis/miniconda/lib/python2.7/site-packages/bs4/__init__.py:166: UserWarning: No parser was explicitly specified, so I'm using the best available HTML parser for this system ("html.parser"). This usually isn't a problem, but if you run this code on another system, or in a different virtual environment, it may use a different parser and behave differently. To get rid of this warning, change this: BeautifulSoup([your markup]) to this: BeautifulSoup([your markup], "html.parser") markup_type=markup_type)) ok | 02 October 2015, 21:53:39 UTC |
9f063db | Luke Yeager | 02 October 2015, 21:01:32 UTC | Merge pull request #324 from gheinrich/dev/torch2 Initial support for Torch7 in DIGITS | 02 October 2015, 21:01:32 UTC |
87f476c | Luke Yeager | 02 October 2015, 18:01:59 UTC | Fix problem with setuptools >= 8.0 - close #337 Add lots of tests, too | 02 October 2015, 18:07:00 UTC |
41dcc78 | Luke Yeager | 01 October 2015, 16:03:55 UTC | Merge pull request #336 from lukeyeager/parse-versions Allow DIGITS to parse NVcaffe version strings | 01 October 2015, 16:03:55 UTC |
cd98383 | Greg Heinrich | 01 October 2015, 12:39:27 UTC | Make only one GPU device visible to Torch | 01 October 2015, 12:39:27 UTC |
935b935 | Greg Heinrich | 04 September 2015, 20:29:22 UTC | Torch integration Unsupported features: - multi-GPU training - generic inference - fine tuning Limitations: - standard networks can only operate with intended image size and number of channels (e.g. LeNet requires grayscale images) | 30 September 2015, 21:24:56 UTC |
45343e7 | Luke Yeager | 30 September 2015, 20:00:02 UTC | Allow DIGITS to parse NVcaffe version strings Previously, it totally failed to parse "0.14.0-alpha", which is valid semantic versioning. | 30 September 2015, 20:00:02 UTC |
7106e1f | Sravan | 04 September 2015, 20:15:54 UTC | Initial revision of Torch-specific files | 30 September 2015, 11:36:49 UTC |
85c51f0 | Luke Yeager | 28 September 2015, 20:04:48 UTC | Merge pull request #333 from lukeyeager/remove-debug-print Remove debug print statement | 28 September 2015, 20:04:48 UTC |
17ca6d9 | Luke Yeager | 28 September 2015, 19:51:55 UTC | Remove debug print statement from #268 | 28 September 2015, 19:51:55 UTC |
f6ac508 | Luke Yeager | 28 September 2015, 19:36:06 UTC | Merge pull request #325 from jmancewicz/file-selector Change file inputs to look the same in chrome and firefox and have bo… | 28 September 2015, 19:36:06 UTC |
95a8548 | Luke Yeager | 28 September 2015, 16:39:45 UTC | Merge pull request #332 from lukeyeager/increase-test-timeouts Increase test timeouts (TravisCI can be slow) | 28 September 2015, 16:39:45 UTC |
7e6f5c8 | Luke Yeager | 28 September 2015, 16:21:55 UTC | Increase test timeouts (TravisCI can be slow) | 28 September 2015, 16:21:55 UTC |
81b3815 | Luke Yeager | 24 September 2015, 17:31:26 UTC | Merge pull request #328 from lukeyeager/change-apt-get-commands Consolidate apt-get commands for easier copy+paste | 24 September 2015, 17:31:26 UTC |
94004e1 | Joe Mancewicz | 22 September 2015, 01:34:37 UTC | Change file inputs to look the same in chrome and firefox and have bootstrap look. | 24 September 2015, 01:12:36 UTC |
6c2738a | Luke Yeager | 23 September 2015, 19:11:10 UTC | Consolidate apt-get commands for easier copy+paste | 23 September 2015, 19:11:10 UTC |
b51e36d | Luke Yeager | 22 September 2015, 21:59:10 UTC | Merge pull request #326 from gheinrich/fix-rounding-issue Fix rounding bug introduced by 80e3330 | 22 September 2015, 21:59:10 UTC |
d278cd5 | Greg Heinrich | 22 September 2015, 21:10:25 UTC | Fix rounding bug introduced by 80e3330 | 22 September 2015, 21:10:25 UTC |
268ab35 | Luke Yeager | 22 September 2015, 20:17:12 UTC | Merge pull request #323 from gheinrich/show_db_backend Show DB backend on home and model creation pages | 22 September 2015, 20:17:12 UTC |
b7e68c0 | Greg Heinrich | 22 September 2015, 19:15:17 UTC | Merge pull request #307 from lukeyeager/pickling-errors Address protobuf PicklingErrors on all platforms | 22 September 2015, 19:15:17 UTC |
f83da34 | Greg Heinrich | 20 September 2015, 20:48:58 UTC | Show DB backend on home and model creation pages | 22 September 2015, 08:20:08 UTC |
0adf025 | Luke Yeager | 18 September 2015, 23:56:54 UTC | Merge pull request #322 from NVIDIA/clarify-cmake-directions Clarify CMake directions | 18 September 2015, 23:56:54 UTC |