Horseshoe Priors for Bayesian Neural Networks.
This software includes an implementation of the model presented in:
"Model Selection in Bayesian Neural Networks via Horseshoe priors", Soumya Ghosh and Finale Doshi-Velez. NIPS 2017. Bayesian Deep Learning Workshop (http://bayesiandeeplearning.org/2017/papers/42.pdf).
The software depends on autograd (https://github.com/HIPS/autograd)
- To train,
- You can use
python ./src/plot_weights.py path/to/results.pkl
to visualize the inferred weights.