Code in this package is meant to be understandable and auditable for correctness. It is not meant to perform especially well.

Distributions includes several optimized modules that require Cython 0.15.1 or later. On Ubuntu 12.04, this can be installed as follows:

sudo apt-get install cython

After making a change to a .pyx or .pxd file, rebuild the extension:

python setup.py build_ext --inplace

To install without cython:

python setup.py --without-cython install

Or:

pip install --global-option --without-cython .

Component models are written as free functions with all data passed in to emphasize their exact dependencies. The distributions.ComponentModel interface class wraps these functions for all models so that they may be used conveniently elsewhere.

The component model functions are strictly divided in to two sets, those which change state and those which do interesting math. State change functions should be simple and fast.

Throughout the interface we use several short variable names:

• ss: sufficient statistics

• hp: hyperparameters

• p: parameters

• y: data value

The functions are as follows:

• create_ss(ss=None, p=None) Create a valid suff stats dict, optionally incorporating the suff stats and parameters provided in the arguments.

• dump_ss(ss) Return a serializable version of the sufficient statistics contained in the argument.

• create_hp(hp=None, p=None) Create a valid hyperparameters dict, optionally incorporating the hps and parameters provided in the argument.

• dump_hp(hp) Return a serializable version of the hyperparameters contained in the argument.

• add_data(ss, y) Add the datapoint to the suff stats. Does not explicitly check that the datapoint is valid. Returns None.

• remove_data(ss, y) Removethe datapoint from the suff stats. Does not explicitly check that the datapoint is valid, or that it was added previously. Returns None.

• sample_data(hp, ss) Sample a datapoint from the distribution described by the hyperparameters and suff stats.

• sample_post(hp, ss) Sample the model parameters from the posterior distribution described by the hyperparameters and suff stats.

• pred_prob(hp, ss, y) Compute the posterior predictive probability of the datapoint, given the hyperparameters and suff stats (and integrating over the model parameters).

• data_prob(hp, ss) Compute the marginal probability of all of the data summarized by the suff stats, given the hyperparameters (and integrating over the model parameters).