That was a toy project for my own learning. it's of zero practical use now. Archived.

A very small automatic differentiation library for C++

You have to install libeigen3-dev first. Support is only for ubuntu

Take a look at main.cpp to read the example. For your own projects, build it, install it, then link ad for your projects.

1. Create a ComputationGraph object.
2. Instantiate the variables. Either from Eigen::MatrixXd that will be copied from, or from shared_ptr to them (like, in parameter). NOTE: The Var objects MUST NOT outlive the ComputationGraph instance they were created from.
3. Do your calculations
4. Backpropagate.

NB: Unlike theano, the ComputationGraph can do only one forward pass. Even if it creates a lot of overhead for "fixed" size models, it makes sequences calculation much much easier to code and to backpropagate into. Look at the example and note how a new graph (which is actually more like a trace of what happened) is created for every new pass.