google/tangent
Source-to-Source Debuggable Derivatives in Pure Python
ArchivedPerforms reverse-mode autodiff by parsing Python source code into an AST, applying symbolic differentiation rules to each node, and emitting readable Python source as output. Supports control flow (loops, conditionals), user-defined functions, and TensorFlow Eager/NumPy operations through a template-based adjoint system. The generated derivatives are fully debuggable and compatible with TensorFlow and NumPy, bridging the gap between ahead-of-time compilation and runtime tracing approaches.
2,316 stars. No commits in the last 6 months.
Stars
2,316
Forks
432
Language
Python
License
Apache-2.0
Category
Last pushed
Sep 29, 2022
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/google/tangent"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Related frameworks
JonathanRaiman/theano_lstm
:microscope: Nano size Theano LSTM module
pranftw/neograd
A deep learning framework created from scratch with Python and NumPy
ahrefs/ocannl
OCANNL: OCaml Compiles Algorithms for Neural Networks Learning
statusfailed/catgrad
a categorical deep learning compiler
mstksg/backprop
Heterogeneous automatic differentiation ("backpropagation") in Haskell