Code accompanying the manuscript "Flexible Context-Driven Sensory Processing in Dynamical Vision Models"
To initialize conda environment, create an empty conda environment with Python 3.12 and pip installed
conda env create --name ei_rnn python=3.12 pip
Alternatively, you can create a conda environment from the requirements/env.yaml file in this repository (Note: there are some extra packages there that you may not need)
conda env create --file requirements/env.yaml --name ei_rnn
This file will create an environment that can be run on a linux machine with a CUDA-capable GPU running drivers compatible with CUDA 12.1 or higher. For other configurations, create the empty environment as above and follow the instructions below for installing the other dependencies.
To install the required packages for your system, activate your environment and pip isntall the file in requirements/ that matches your system
conda activate ei_rnn
pip install -r requirements/<requirements_file>
where <requirements_file> should be replaced with cu117.txt, cu121.txt, linux_cpu.txt, osx.txt based on your system
To train the model, execute the following command in your active conda environment
python train.py <kwargs>
You should replace with <kwargs> with any parameters you wish to override. The defaults can be found in config/config.yaml. Leave this blank if you wish to use the defaults