/QuickBNN

Quick Construction of BNNs using Flux like syntax and Turing as a backend

Primary LanguageJulia

QuickBNN

Quick Construction of BNNs using Flux like syntax and Turing as a backend

QuickBNN is a work in progress. It allows the construction of genereal Baysian Neural Networks following syntax similar to Flux. Turing is used as an inferential backend.

Example

using Turing
using Flux
using QuickBNN

ar1 = AR([0.5])
y = ar1(;N=500)
x = y[1:end-1]
y = y[2:end]

net = ChainBNN(DenseBNN(1, 2, :sigmoid), DenseBNN(2, 1))
bnn = BNN(net)
model = bnn(y, hcat(x...))
chain = sample(model, NUTS(), MCMCThreads(), 10_000, 4)
plot(chain)