I am currently exploring Bayesian Neural Network application on timeseries and stumbled on pymc3 library. But don't exactly understand how can I use it on a timeseries data.

I am coming from a background of using statistical models:ARIMA, GARCH on timeseries.

To start I want to implement a simple Bayesian feed-forward neural network on a timeseries data.

What I am thinking to do is to set AR & MA values of a univariate timeseries as the priors for my model.Therefore analyze their distribution to build my posterior.

Could someone guide is it logical to start this way?


closed as too broad by Firebug, kjetil b halvorsen, jbowman, Stephan Kolassa, AdamO Dec 13 '17 at 17:40

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.


Usually there are two ways to model time-series in Deep Learning:

  1. Recurrent Neural Networks like LSTM (there are many resources if you google). You should be able to implement those in PyMC3, especially if they are supported by Lasagne.
  2. Placing a stochastic process like random-walk or ARMA on the parameters of your neural network. See my blog post for more info: http://twiecki.github.io/blog/2017/03/14/random-walk-deep-net/

Not the answer you're looking for? Browse other questions tagged or ask your own question.