Sharing: Tsetline machine implementation based on tensor operations

cair / TsetlinMachine

Code and datasets for the Tsetlin Machine

https://arxiv.org/abs/1804.01508

MIT License

466 stars 51 forks source link

Sharing: Tsetline machine implementation based on tensor operations #14

Closed fwaris closed 2 years ago

fwaris commented 2 years ago

Please see: https://github.com/fwaris/FsTsetlin

(Still very much WIP)

The core of the Tsetlin machine is implemented using tensor operation (which can be parallelized).

The front-end language is F# (.Net) but it uses the tensor library underpinning PyTorch (via TorchSharp)

Trains and predicts on GPU/CPU.

[edited] Have to say the original Tsetlin machine paper is very well written. I wrote almost the entire code just from the paper.

olegranmo commented 2 years ago

Looks very interesting and thanks for the kind feedback on the paper! Look forward to trying it out :-)

fwaris commented 2 years ago

did some quick tests with the "noisy xor" dataset to compare CPU vs. GPU performance. CPU is faster with 10 clauses (per class). As clauses are increased steadily to 1000, CPU performance degrades quickly while GPU performance remains quite steady. This is not exhaustive testing but shows that system is working as intended.