Iterating through new ideas. Seeing how they work out. Then start all over again. Get a spike of happiness if it works.
Nengo implementation of reinforcement learning in watermaze tasks using mouses. Intended to be run on the Loihi chip as well.
This project is intended to be run using NxSDK 0.9, which requires Python 3.5.2.
The nengo-loihi installation page suggests using miniconda, but unfortunately miniconda does not actually contain 3.5.2 in its repository (not for Arch at least).
So for me the following (using mini conda) does not resolve, because the package repositories do not contain the correct version:
conda create --name loihi python=3.5.2
So at this point we should resort to building Python 3.5.2 from source.
But since I don't care about maintaining a global Python 3.5.2 install, I instead checked out pyenv which does contain it:
pyenv install --list | grep 3.5.2
3.5.2
anaconda3-5.2.0
pypy3.3-5.2-alpha1-src
pypy3.3-5.2-alpha1
pyenv allows us to activate Python 3.5.2 only for this project and will do the compiling for us.
Install with:
pyenv install -v 3.5.2
N.B. my first build failed due to the correct version of OpenSLL not being found. I had a look at common build problems for pyenv and followed instructions for my system (Arch). I double checked I installed all dependencies and also made sure to point to the correct (older) version of OpenSLL. Long ive downgrading.
LDFLAGS="-L/usr/lib/openssl-1.0" \
CFLAGS="-I/usr/include/openssl-1.0"\
pyenv install -v 3.5.2
Setting LDFLAGS and CFLAGS was necessary for successful compilation for me.
Activate as your global python version using:
pyenv global 3.5.2
(Go back with to the default system python installation using: pyenv global system).
Or only use 3.5.2 locally (recommended; assumes you are in our NMC project repo):
pyenv local 3.5.2
This creates a .python-version file.
From here on you can install packages with your preferred method, either straight into the 3.5.2. installation with pip, or in some virtual environment.
To ensure python indeed refers to the one selected with pyenv, add the following to your .bashrc or .zshrc:
export PYENV_ROOT="$HOME/.pyenv"
export PATH="$PYENV_ROOT/bin:$PATH"
eval "$(pyenv init -)"