This repository houses a final project for the Fall 2018 Harvard APMTH207 course. We selected two papers related to Bayesian artificial neural networks and made an exploratory and didactic tutorial, in the jupyter notebook BBB_tutorial.ipynb
. More techically, we test the performance of ADVI in pymc3
using the Fashion-MINIST dataset and an active learning task on this dataset. Note that in the /staging
folder we have older versions of the notebooks that may also provide useful API references. We include instructions below to help get the code running:
To make the environment here use the following (assumes anaconda installation):
conda env create -f environment.yml
And, if you're worried about system pollution, it can just as easily be removed with:
conda env remove -n bbb_models
To install additional packages to the environment simply conda activate bbb_models
to use the environment and use conda or pip to install as usual. NOTE: If using jupyter notebook or jupyterlab use the following to add the kernel to your options:
conda install -n bbb_models nb_conda_kernels
This doesn't need to be done in the bbb_models
environment, just the one from which you launch jupyterlab or jupyter notebook.