Extend to support recursive sequential

ChenAo-Phys / pytorch-Jacobian

Implement efficient jacobian method available for most pytorch networks

MIT License

42 stars 4 forks source link

Extend to support recursive sequential #1

Open nodtem66 opened 2 years ago

nodtem66 commented 2 years ago

I have used this library to calculate jacobians for the Neural tangent kernel study. It's fast and reliable and I'm just thinking to extend the library for recursive submodule by modifying extend() with some code that is used to iterate the computational graph, e.g. torchinfo.

Is it possible? I just want to make sure before wasting time doing it. Thank you :D

ChenAo-Phys commented 2 years ago

Yea sure! I'm glad to hear that my program helps you and you would like to extend it. Thanks in advance :)