What are your goals for this club?

Here’s mine

  • Deeply understand state space models from an inference and estimation persective
  • Be able to read the mathematical notation surrounding state space models with ease
  • Be able to code various types of State Space models easily and fluently with Dynamax in particular
  • Learn how other folks are using SSMs in their own work

Similar to yours above, but in particular i’d like to see more applied examples. Much of the reading i’ve done had gone through examples on how to estimate the model and detailed the mechanics of the model, but there has been less focus on how to convert e.g. the estimated latent states of a hidden markov model into actionable insights to inform a decision.

I think i even have a good use case for these kinds of model in work as i have a big system which has some observable components and proxy variables and risk profiles differ depending on the state of the system… but say i estimate a model which is a good fit to historic data… what are the recipes for abstracting the covariate settings which move the system into riskier states say…

Not sure if i’ve expressed myself well here. But the broad thought is : how do i convert a well estimated state space model into a compelling case for action or decision?

  • Better understand the use cases for the more advanced techniques (other than linear gaussian)
  • Hear some practical examples for SSMs beyond time series forecasting
  • Familiarise myself with dynamax + sts-jax (and maybe how these can be used in larger jax-based ppls)
1 Like

On the applied side:

  • get to learn more jax and assorted libraries
  • learn more about modelling techniques

and on the mathematical side:

  • I am mostly familiar with (Extended) Kalman Filter/ Smoother and some simulation based approaches (particle filters/importance sampling), so I’m interested in the algorithms that dynamax uses (Gaussian Hermite KF?)

My goals are to:

  • Learn more about jax and dynamax
  • Find ways to apply SSMs in my work in ecosystem modeling

My goals are:

  • To better understand advanced Bayesian models.
  • To explore modern coding/development techniques.
  • Having a long term problem to solve with the tools and techniques learnt here.
  • Fluency with jax - especially, translating math code to jax
  • Understand SSM especially non-linear gaussians
  • Understand particle filters
  • Implement a few from scratch to play around with dynamics
  • Fluency in expressing diverse state space models in python

Using this as a forcing function to help learn more about state space models and work up to structural time series and causal impact

1 Like