Tf Explain Versions Save

Interpretability Methods for tf.keras models with Tensorflow 2.x

v0.3.1

2 years ago

What's Changed

New Contributors

Full Changelog: https://github.com/sicara/tf-explain/compare/v0.3.0...v0.3.1

v0.3.0

3 years ago

Release includes:

  • Migration from TravisCI to Github Actions
  • Compatibility with TensorFlow 2.3 and 2.4
  • Fix in Vanilla Gradients computation (use raw scores instead of softmax outputs)
  • Let user choose between opencv versions

v0.2.1

4 years ago

In this release:

  • Checked against TensorFlow 2.1.0
  • image_weight parameter added to GradCAM to better handle visualizations
  • Grids of attribution maps are now normalized image by image instead of the whole grid
  • bugfix : reshape of non-square visualization by @jpsimen

v0.2.0

4 years ago

In this release:

  • Gradients*Inputs algorithm by @ywolff
  • Add layer detection for Grad CAM (last convolutional output) to ease user experience with this method
  • Better handling of image output ([0, 1], [-1, 1] & [0 255])
  • Notebooks

v0.1.0

4 years ago

In this release:

  • TensorFlow 2.0.0 compatibility
  • Add Vanilla Gradients implementation
  • Add Integrated Gradients implementation
  • Take Viridis as default colormap

Thanks to @ywolff for the contributions in this release.

v0.0.2-alpha

4 years ago

In this release:

  • Add SmoothGrad implementation
  • Improve computation time with tf.function decorator
  • Tensorboard integration
  • Docstring & Testing improvements

0.0.1-alpha

4 years ago

First version of the package, with 2 different APIs:

  • a core API to use on loaded models, outside of training
  • a callback API to use inside .fit and .fit_generator methods

Available methods in this release are:

  • Activations Visualization
  • Occlusion Sensitivity
  • Grad CAM