TensorFlow Addons v0.13.0
Release Notes
- Built against TensorFlow 2.5
- CUDA kernels are compiled with CUDA11.2 and cuDNN 8.1.0
- API docs found on the website
Changelog
tfa.activations
- Cleanup legacy codes for activations (#2394)
tfa.image
- Add python fallback for adjust_hsv_in_yiq (#2392)
- Remove ImageProjectiveTransform kernel (#2395)
- Fix EDT float16 and float64 kernels (#2412)
- Optimize EDT (#2402)
- Update cutout_ops.py (#2416)
tfa.metrics
- Initial commit of streaming Kendall's Tau algorithm. (#2423)
- Fix F1Score docs (#2462)
- Matthew Fix (#2406)
- Fix RSquare serialization (#2390)
- Make RSquare.reset_states to be able to run in tf.function (#2445)
tfa.optimizers
- Adding COntinuos COin Betting (COCOB) Backprop optimizer (#2063)
- Fix NovoGrad optimizer to work with float64 layers (#2467)
- Update cyclical_learning_rate.py (#2286)
- RectifiedAdam: Store 'total_steps' hyperparameter as float (#2369)
tfa.text
- fix wrong type hinting of
crf_log_likelihood
(#2471)
Thanks to our Contributors
@0x0badc0de, @bhack, @DragonPG2000, @Harsh188, @WindQAQ, @ashutosh1919, @fsx950223, @jeongukjae, @jonpsy, @juliangilbey, @lucasdavid, @lum4chi, @m-a-r-o-u, @nickswalker, @nleastaugh, @npanpaliya, @olesalscheider, @rehanguha, @seanpmorgan, @shubhanshu02, @sorensenjs, @whatwilliam and @xiedeping