Pipable plug-and-play inference methods for time series analysis with state space models.

cat theta.json | ./simplex -M 10000 -P | ./ksimplex -M 10000 -P > mle.json
cat mle.json | ./kmcmc -M 100000 --full -P | ./pmcmc -J 1000 -M 500000 --full

This README provided information for developers or users wanting to build pipable inference methods on their local machine. If you want to use those methods, see plom.io

plom.io provides a compilation service that generate standalone binaries that you can run on your machine.

All the methods are implemented in plain C. The C code contain generic part (working with any models) and model specific part. The specific parts are templated using Python.

##Dependencies

C:

• gsl: http://www.gnu.org/software/gsl/
• zmq: http://www.zeromq.org/
• jansson: http://www.digip.org/jansson/
• openMP: http://openmp.org/ (openMP is optional as it is not supported by Clang. All the feature provided by openMP have a counterpart provided by zmq and pthread.)

Python:

• Python 2.7.x: www.python.org/
• Django: https://www.djangoproject.com/
• SymPy: http://sympy.org/

##Building the C libraries in model_builder/C:

make
make install

##Creating and installing the python package

At the root of the repo run:

python setup.py sdist

in the package directory:

python setup.py install

##Generating the model-specific code:

From the command line, run:

pmbuilder context.json process.json link.json -o path_model_coded_in_C

In your script you can use:

import json
from plom.Builder import PlomModelBuilder

c = json.load(open('example/noise/context.json'))
p = json.load(open('example/noise/process.json'))
l = json.load(open('example/noise/link.json'))

model = PlomModelBuilder('path_model_coded_in_C', c, p, l)

model.prepare()
model.write_settings()
model.render()

##Building the inference methods

in path_model_coded_in_C/C/templates:

make
make install

All the inference methods binaries are now available in path_model_coded_in_C

On model_builder/doc/ run: doxygen Doxyfile and open model_builder/doc/html/index.html with a web browser.

After having learned the basic structures involved in core, we recommend to use call and caller graphs as an entry point to the sources.

In model_builder:

python -m unittest discover

GPL version 3 or any later version.

We want to thank:

• Professor Bryan Grenfell.

• Professor Bernard Cazelles.

• Princeton University and the "simforence" project for letting us release the sources under the GPLv3 license.