INSTALL.md

How to install pangolin

Step 1 (Optional but recommended) Install Anaconda and create a new environment, then activate it.

conda create -n pangolin-test python=3.12
conda activate pangolin-test

(If you are a Python expert, you could also use any other Python virtual environment tools, but Python 3.11 or higher is required.)

Step 2 Download pangolin source code and extract it. You could do this in either of the following ways:

1. Go to https://github.com/justindomke/pangolin in your browser, then click on "Code" and then "Download zip". Then double-click the file to extract it and past the contents wherever you want to store them.

2. Type the following at the command line (assumes you have wget and unzip installed):

   cd path/to/wherever/
   wget https://github.com/justindomke/pangolin/archive/refs/heads/main.zip
   unzip main.zip
   

Step 3. Install required packages. At the command line, go to wherever you put the pangolin source code, and type the following:

pip install -r pangolin-main/requirements.txt

Step 4. Install pangolin from local directory. At the command line, go to wherever you put the pangolin source code, and type the following:

pip install pangolin-main/

At this point, technically you're done. But let's test to make sure installation worked.

Step 5. Start Python make sure Pangolin works.

% python
Python 3.12.4
Type "help", "copyright", "credits" or "license" for more information.
>>> import pangolin as pg
>>> x = pg.normal(0,1)
>>> pg.print_upstream(x)

You should see something like this:

shape | statement
----- | ---------
()    | a = 0
()    | b = 1
()    | c ~ normal(a,b)

Step 6. Make sure that Pangolin's dependencies (Jax and NumPyro) are correctly installed. Start Python and do:

import numpyro
import jax

def model():
    x = numpyro.sample('x',numpyro.distributions.Normal(0,1))
    
mcmc = numpyro.infer.MCMC(numpyro.infer.NUTS(model), num_warmup=100, num_samples=100)
mcmc.run(jax.random.PRNGKey(0))

You should see some output like

sample: 100%|██████████| 200/200 [00:01<00:00, 123.72it/s, 3 steps of size 1.40e+00. acc. prob=0.84]