JSL is a JAX library for Bayesian inference in state space models. This is an unstable work in progress, and will eventually be replaced by ssm-jax. However, as of 2022-05-16, you should continue to use JSL.
We assume you have already installed JAX and Tensorflow, since the details on how to do this depend on whether you have a CPU, GPU, etc. (This step is not necessary in Colab.)
Now install these packages:
!pip install --upgrade git+https://github.com/google/flax.git
!pip install --upgrade tensorflow-probability
!pip install git+https://github.com/blackjax-devs/blackjax.git
!pip install git+https://github.com/deepmind/distrax.git
Then install JSL:
!pip install git+https://github.com/probml/jsl
Alternatively, you can clone the repo locally, into say ~/github/JSL, and then install it as a package, as follows:
!git clone https://github.com/probml/JSL.git
cd JSL
!pip install -e .
You can see how to use the library by looking at some of the demos. You can run the demos from inside a notebook like this
%run JSL/jsl/demos/kf_tracking.py
%run JSL/jsl/demos/hmm_casino_em_train.py
Or from inside an ipython shell like this
from jsl.demos import kf_tracking
figdict = kf_tracking.main()
Most of the demos create figures. If you want to save them (in both png and pdf format), you need to specify the FIGDIR environment variable, like this:
import os
os.environ["FIGDIR"]='/Users/kpmurphy/figures'
from jsl.demos.plot_utils import savefig
savefig(figdict)
Gerardo Durán-Martín (@gerdm), Aleyna Kara(@karalleyna), Kevin Murphy (@murphyk), Giles Harper-Donnelly (@gileshd), Peter Chang (@petergchang).