Commits · a2052cad29242fd27acea167900c438ed47eda81 · Sciencebeam / Sciencebeam Gym · GitLab

Snippets Groups Projects

Dec 12, 2017
- Merge pull request #7 from elifesciences/filter-records-by-annotation-being-present · a2052cad
  Daniel Ecer authored 7 years ago
```
filter records by annotation being present
```
  a2052cad
- filter records by annotation being present · 5abab19f
  Daniel Ecer authored 7 years ago
  
  5abab19f
- Merge pull request #6 from elifesciences/filter-records-by-page-no · ecdbfdd4
  Daniel Ecer authored 7 years ago
```
Filter training records by page no
```
  ecdbfdd4
- added workaround for tensorflow 1.2 · 8abe7984
  Daniel Ecer authored 7 years ago
  
  8abe7984
- allow page range to be specified for model training · c73a3a3a
  Daniel Ecer authored 7 years ago
  
  c73a3a3a
- moved parsing examples to read_examples; removed defunct batch_of_examples (no metrics-updates op) · ac236b21
  Daniel Ecer authored 7 years ago
  
  ac236b21
Dec 11, 2017
- moved reading examples to tf.data api · 26bbec5f
  Daniel Ecer authored 7 years ago
  
  26bbec5f
- Merge pull request #5 from elifesciences/add-support-for-class-weights · ee193b5b
  Daniel Ecer authored 7 years ago
```
add support for class weights
```
  ee193b5b
- changed tf runtime version in eval script to match train script · 2c0e32fd
  Daniel Ecer authored 7 years ago
  
  2c0e32fd
- added missing arguments to eval script · 327d399a
  Daniel Ecer authored 7 years ago
  
  327d399a
- added print statement before each tfrecords section · 8e2117fb
  Daniel Ecer authored 7 years ago
  
  8e2117fb
- added weighted sample weighted cross entropy loss · 07e040ac
  Daniel Ecer authored 7 years ago
  
  07e040ac
Dec 08, 2017
- optionally print debug information (pos_weight) · d424348e
  Daniel Ecer authored 7 years ago
  
  d424348e
- added page no to uri in tfrecord and added separated page no field · 9ac3e055
  Daniel Ecer authored 7 years ago
  
  9ac3e055
- added option to calculate sample class weights (instead of using same class... · 3732ba26
  Daniel Ecer authored 7 years ago
```
added option to calculate sample class weights (instead of using same class weights for all samples)
```
  3732ba26
- increased google ml runtime environment to 1.2 · e7c3e449
  Daniel Ecer authored 7 years ago
  
  e7c3e449
Dec 07, 2017
- added test to confirm that loss function supports sample based weights · 5432f53a
  Daniel Ecer authored 7 years ago
  
  5432f53a
- use default unknown weight and get unknown weight from class weights if present · 71081bfd
  Daniel Ecer authored 7 years ago
  
  71081bfd
- removed limited and unnecessary override_if_not_in_args · d05fc08e
  Daniel Ecer authored 7 years ago
  
  d05fc08e
- don't swallow argument errors · b0f73db8
  Daniel Ecer authored 7 years ago
  
  b0f73db8
- added MAX_TRAIN_STEPS script configuration; fixed passing in CLASS_WEIGHTS_FILENAME · 6d7a715d
  Daniel Ecer authored 7 years ago
  
  6d7a715d
- optionally include unknown in class weights · 7bf05aff
  Daniel Ecer authored 7 years ago
  
  7bf05aff
- don't include labels by default with a zero class weight · 4400522a
  Daniel Ecer authored 7 years ago
  
  4400522a
- train with class weights · c5f49894
  Daniel Ecer authored 7 years ago
  
  c5f49894
- marked slow tests · 71c8b056
  Daniel Ecer authored 7 years ago
  
  71c8b056
Dec 06, 2017
- added simple model test · 53bc6767
  Daniel Ecer authored 7 years ago
  
  53bc6767
- moved loss functions to separate module · 3cc6870c
  Daniel Ecer authored 7 years ago
  
  3cc6870c
- added tqdm dependency · e27ab64f
  Daniel Ecer authored 7 years ago
  
  e27ab64f
- calculate median balanced class weights · c9542bc8
  Daniel Ecer authored 7 years ago
  
  c9542bc8
- allow separate training and data suffix · 20aafa12
  Daniel Ecer authored 7 years ago
  
  20aafa12
Dec 05, 2017
- add dataset suffix to default job id · 7d75e3d7
  Daniel Ecer authored 7 years ago
  
  7d75e3d7
- added build status and license badge · 43c8a2ae
  Daniel Ecer authored 7 years ago
  
  43c8a2ae
- use GroupTransforms for other transforms · b815829d
  Daniel Ecer authored 7 years ago
  
  b815829d
- make number of workers configurable (script) · e7e163ea
  Daniel Ecer authored 7 years ago
  
  e7e163ea
- added PreventFusion transform to better distribute load across workers · 17c5157e
  Daniel Ecer authored 7 years ago
  
  17c5157e
Dec 04, 2017
- count file pairs and pages while preprocessing · 9bc7ff91
  Daniel Ecer authored 7 years ago
  
  9bc7ff91
- added additional preprocessing job name suffix to better be able to distinguish jobs · f4f9ebe6
  Daniel Ecer authored 7 years ago
  
  f4f9ebe6
- minor change to default preproc path · 86c326e5
  Daniel Ecer authored 7 years ago
  
  86c326e5
- Merge pull request #3 from elifesciences/full-training-pipeline · ede6799e
  Daniel Ecer authored 7 years ago
```
Full preprocessing pipeline
```
  ede6799e
- added dependencies section to README · c28b3853
  Daniel Ecer authored 7 years ago
  
  c28b3853