Please sign in to comment.
Sound Classifier: Allow Reusing Deep Features (#1657)
* Sound Classifier: Allow Reusing Deep Features - Add tc.sound_classifier.get_deep_features. - Change sound classifier (create & predict) to use get_deep_features. - Allow sound classifier (create & predict) to use deep features directly. - Use SArrayBuilder for deep features - Logging: print header info only if printing progress info. - Any function which could result in deep feature extraction should have `verbose` flag. - Allow predict to do parameter checking. Rather than doing not as well in a bunch of other places that just call predict - classify/evaluate/predict_topk should take SFrame or dict in addition to SArray. - Unit tests: test deep feature reuse. - Unit tests: copy data rather than regenerate each time. - Unit tests: make problematic test less sensitive (six decimal places is fine). - Unit tests: test validation set actually getting used or not. - Took a first stab at the user guide section for this feature, more will be added later (#1649). - Fixed finicky unit test
- Loading branch information...
Showing with 298 additions and 96 deletions.
- +65 −5 src/unity/python/turicreate/test/test_audio_functionality.py
- +1 −1 src/unity/python/turicreate/toolkits/audio_analysis/audio_analysis.py
- +1 −1 src/unity/python/turicreate/toolkits/sound_classifier/__init__.py
- +41 −12 src/unity/python/turicreate/toolkits/sound_classifier/_audio_feature_extractor.py
- +161 −77 src/unity/python/turicreate/toolkits/sound_classifier/sound_classifier.py
- +29 −0 userguide/sound_classifier/advanced-usage.md
Oops, something went wrong.