There is a newer version of this record available.

Software Open Access

datascienceinc/Skater: Skater 1.1.1-b2

Aaron Kramer; Pramit Choudhary; Ben Van Dyke; Alvin Thai; Nitin Pasumarthy; Guillaume Lemaitre; Dave Thompson

New Features:

  • Added new interface skater.core.local_interpretation.dnni.deep_interpreter.DeepInterpreter for interpreting tensorflow and Keras based models
  • enabling support for interpreting DNNs using gradient-based e-LRP and Integrated Gradient through DeepInterpreter.explain
  • added support to visualize relevance/attribution scores for interpreting image and text inputs
    • skater.core.visualizer.image_relevance_visualizer.visualize
    • skater.core.visualizer.text_relevance_visualizer import build_visual_explainer, show_in_notebook
  • user-friendly Utility functions to generate simple yet effective conditional adversarial examples for image inputs
    • skater.util.image_ops import load_image, show_image, normalize, add_noise, flip_pixels, image_transformation
    • skater.util.image_ops import in_between, greater_than, greater_than_or_equal
  • More interactive notebook use-cases for building and interpreting DNNs for evaluating model stability/identifying blind spots
  • Updates to documentation, https://datascienceinc.github.io/Skater/overview.html
  • New section summarizing Notebook examples https://datascienceinc.github.io/Skater/gallery.html
  • Other bug fixes Credits:
  • Special thanks to Marco Ancona(@marcoancona) for guiding in enabling this feature within Skater.
  • Thanks to all other contributors for helping move the library forward every day.

Files (50.1 MB)
Name Size
datascienceinc/Skater-v1.1.1-b2.zip
md5:c508be88b7b6e309370d65d51c0f76aa
50.1 MB Download
643
65
views
downloads
All versions This version
Views 64367
Downloads 6514
Data volume 9.1 GB701.3 MB
Unique views 53865
Unique downloads 3410

Share

Cite as