| Vijay Vasudevan | 10e62dc | 2015-12-11 23:03:16 -0800 | [diff] [blame^] | 1 | # Changes since last release |
| 2 | |
| 3 | ## Breaking changes to the API |
| 4 | |
| 5 | * `AdjustContrast` kernel deprecated, new kernel `AdjustContrastv2` takes and |
| 6 | outputs float only. `adjust_contrast` now takes all data types. |
| 7 | * `adjust_brightness`'s `delta` argument is now always assumed to be in `[0,1]` |
| 8 | (as is the norm for images in floating point formats), independent of the |
| 9 | data type of the input image. |
| 10 | * The image processing ops do not take `min` and `max` inputs any more, casting |
| 11 | safety is handled by `saturate_cast`, which makes sure over- and underflows |
| 12 | are handled before casting to data types with smaller ranges. |
| 13 | |
| Vijay Vasudevan | 2c3738d | 2015-12-08 14:55:13 -0800 | [diff] [blame] | 14 | # Release 0.6.0 |
| 15 | |
| 16 | ## Major Features and Improvements |
| 17 | |
| 18 | * Python 3.3+ support via changes to python codebase and ability |
| 19 | to specify python version via ./configure. |
| 20 | |
| 21 | * Some improvements to GPU performance and memory usage: |
| 22 | [convnet benchmarks](https://github.com/soumith/convnet-benchmarks/issues/66) |
| 23 | roughly equivalent with native cudnn v2 performance. Improvements mostly due |
| 24 | to moving to 32-bit indices, faster shuffling kernels. More improvements to |
| 25 | come in later releases. |
| 26 | |
| 27 | |
| 28 | ## Bug fixes |
| 29 | |
| 30 | * Lots of fixes to documentation and tutorials, many contributed |
| 31 | by the public. |
| 32 | |
| 33 | * 271 closed issues on github issues. |
| 34 | |
| 35 | ## Backwards-incompatible changes |
| 36 | |
| 37 | * tf.nn.fixed_unigram_candidate_sampler changed its default 'distortion' |
| 38 | attribute from 0.0 to 1.0. This was a bug in the original release |
| 39 | that is now fixed. |
| 40 | |
| Vijay Vasudevan | ddd4aaf | 2015-12-08 09:58:59 -0800 | [diff] [blame] | 41 | # Release 0.5.0 |
| 42 | |
| 43 | Initial release of TensorFlow. |