Deephy

Latest version: v0.5.3

Safety actively analyzes 623608 Python packages for vulnerabilities to keep your Python projects secure.

Scan your dependencies

Page 1 of 5

1.15.0

<!--- https://github.com/mgroth0/deephy/releases -->

[//]: (VERSION:1.15.0)


Compatible pip package
version: [0.5.3](https://pypi.org/project/deephy/0.5.3/) ([instructions](https://colab.research.google.com/drive/1HAaVOopHDNVKryP14wW4K_rcqeeqYrLK#scrollTo=VtUgz8xGYKHj))

PIP Python Package Updated to 0.5.3
- Added `Stopwatch` helper class
- Updated from `cbor` pip package to `cbor2` which is much better
- Saves in the new formats described below
- Added convenience function `import_torch_dataset` to pip package (see instructions above)
- Creating an `ImageFile` directly now will be more difficult. If you cannot migrate to `import_torch_dataset`, please let me know

[//]: ( New Features)


Performance Improvements

- Changed pixel format from float64 to int8
- Changed activation format from float64 to float32
- Saving `Test`data from using the pip package should be 6-10 times faster
- `.test` files should be 7-8 times smaller
- Loading tests in app should be faster
- Clicking images in app should be faster



[//]: ( Cosmetic Changes)

[//]: ( Bug Fixes)


[//]: ( Notes)

[//]: ( Todo)

1.14.1

<!--- https://github.com/mgroth0/deephy/releases -->

[//]: (VERSION:1.14.1)


Compatible pip package
version: [0.4.0](https://pypi.org/project/deephy/0.4.0/) ([instructions](https://colab.research.google.com/drive/1PNiGD26uBsktq64fqPg76yoN-ruixavj))

[//]: ( New Features)

Performance Improvements

- Small optimization with a large effect on performance when clicking images

[//]: ( Cosmetic Changes)

[//]: ( Bug Fixes)


[//]: ( Notes)

[//]: ( Todo)

1.14.0

<!--- https://github.com/mgroth0/deephy/releases -->

[//]: (VERSION:1.14.0)


Compatible pip package
version: [0.4.0](https://pypi.org/project/deephy/0.4.0/) ([instructions](https://colab.research.google.com/drive/1PNiGD26uBsktq64fqPg76yoN-ruixavj))

New Features

- functional "Forward" and "Back" buttons for the ByImage view

[//]: ( Performance Improvements)


[//]: ( Cosmetic Changes)

[//]: ( Bug Fixes)


[//]: ( Notes)

[//]: ( Todo)

1.13.0

<!--- https://github.com/mgroth0/deephy/releases -->

[//]: (VERSION:1.13.0)


Compatible pip package
version: [0.4.0](https://pypi.org/project/deephy/0.4.0/) ([instructions](https://colab.research.google.com/drive/1PNiGD26uBsktq64fqPg76yoN-ruixavj))

New Features

- Add activation ratios for top neurons in bound test viewers

[//]: ( Performance Improvements)


[//]: ( Cosmetic Changes)

[//]: ( Bug Fixes)


[//]: ( Notes)

[//]: ( Todo)

1.12.0

<!--- https://github.com/mgroth0/deephy/releases -->

[//]: (VERSION:1.12.0)


Compatible pip package
version: [0.4.0](https://pypi.org/project/deephy/0.4.0/) ([instructions](https://colab.research.google.com/drive/1PNiGD26uBsktq64fqPg76yoN-ruixavj))


New Features
- If ", texture :" is detected in category name, only the text found be fore the comma is displayed in the predictions list

Performance Improvements

- Set up a new asynchronous loading system, improving performance throughout app
- Reduced RAM usage by using floats instead of doubles
- Implemented Multik to get maximum neuron activations much faster


Cosmetic Changes
- Greatly improved the look of the text and spacing in the ByImage view. Some of the information is now truncated or rounded again, but it can all be seen by mousing over the text and looking at the tooltip

Bug Fixes
- top neuron sorting now properly follows the normalization setting

[//]: ( Notes)

[//]: ( Todo)

1.11.2

<!--- https://github.com/mgroth0/deephy/releases -->

[//]: (VERSION:1.11.2)


Compatible pip package
version: [0.4.0](https://pypi.org/project/deephy/0.4.0/) ([instructions](https://colab.research.google.com/drive/1PNiGD26uBsktq64fqPg76yoN-ruixavj))


[//]: ( New Features)

Cosmetic Changes

- Remove significant figures function from prediction values, to look more precise

Bug Fixes

- activation values in ByImage view now normalize in the proper way

[//]: ( Notes)

[//]: ( Todo)

Page 1 of 5

© 2024 Safety CLI Cybersecurity Inc. All Rights Reserved.