Original Paper: Learning to Communicate: Channel Auto-encoders, Domain Specific Regularizers, and Attention
-
Clone the repo and cd into it.
git clone https://github.com/bitsandscraps/learn2com.git cd learn2com
-
Install the virtual evironment. I use
pipenv
but you may use whatever program you want.pipenv --three
To use
pipenv
, you need to first erase the line that hastensorflow
in the Pipfile, and then re-installtensorflow
in any way you want.
pipenv run python -m learn2com.main
learn2com.main
can take a handful of command-line options.
Check it out using the -h
option, or reading through the code.
- The delay channel is not working properly. The network fails miserably with delay regularization.
- The performance is not even close to what the paper is saying.
- Plot-generation code
- Localization networks