jupyterstan

jupyterstan is a package to help development of Stan models (using pystan) in jupyter notebooks.

The package is heavily based on Arvinds-ds stanmagic package, but provides an interface that simply returns a pystan.Model object.

In addition, it bundles Arvinds-ds stan_code_helper package to improve syntax highlighting for stan cells.

Features

• Stan language syntax highlighting in all cells beginning with %%stan
• Compile a stan model and save it as a pystan variable by running a %%stan cell
• No longer worry about model_code, reading in stan files, etc.
• Support for caching compiled stan models: re-executing a cell with the same code will not recompile the code but uses a cached version.

Installation

To install the library:

pip install jupyterstan

Usage

To use the magic in your notebook, you need to lead the extension:

%load_ext jupyterstan

To define a stan model inside a jupyter notebook, start a cell with the %%stan magic. You can also provide a variable name, which is the variable name that the pystan.Model object will be assigned to. For example:

%%stan paris_female_births
data {
    int male;
    int female;
}

parameters {
    real<lower=0, upper=1> p;
}

model {
    female ~ binomial(male + female, p);
}

When you run this cell, jupyterstan will create a pystan Model object, which will compile your model and allow you to sample from it. To use your compiled model:

fit = paris_female_births.sampling(
    data={'male': 251527, 'female': 241945},
    iter=1000,
    chains=4
)