A PyTorch implementation of Unbiased Online Recurrent Optimization; Corentin Tallec, Yann Ollivier.
The main file to run is demo_uoro/demo_uoro_anbn.py. This compares UORO and RTRL on the a^n b^n dataset.
To set up this repo, go:
git clone https://github.com/petered/uoro-demo.git
cd uoro-demo
source setup.sh
(You'll be prompted to setup pytorch manually depending on your system).
Or to be able to import this code in an existing Python environment, go:
pip install -e git+https://github.com/petered/uoro-demo.git#egg=uoro-demo
You'll need the artemis dependency to be checked out on branch ("peter") for now.
This repo has been auto-generated by the Public Release Package