Bayesian analysis of SiV spectral data and spectroscopic temperature determination in python.
Author: Ruffin Evans (ruffinevans@gmail.com)
Contains contributions from yaml-serialize by Tom Dimiduk.
Example data provided by Christian Nguyen with contributions from the rest of the Lukin Group SiV team in the Harvard Physics Department.
This package is designed to analyze fluorescence spectra from SiV centers in diamond by performing Bayesian parameter estimation. The emphasis is on ease-of-use and producing reliable estimates of the fluorescence wavelength and associated uncertainty.
The package is designed to fit one of two possible models to the data:
- A model containing a single Lorentzian peak describing the SiV center fluorescence
- A model containing two Lorentzian peaks an experimental peak and a reference/calibration peak.
Although the second model is more complicated, it allows an absolute frequency calibration with every spectrum, removing several sources of possible systematic error.
Simply download the package and run python setup.py install within the package directory.
To test that the package is performing as intended, you can run the nosetests command (or python setup.py test). These tests will test some basic functionality of the package, including performing some simple MC-based inference on some test data. This inference should only take a small amount of time, probably less than a minute on your machine. However, please be patient while these tests run. If they hang for more than a few minutes, something is probably wrong -- please file a bug report.
Currently, all of the documentation is in three jupyter notebooks:
tutorial.ipynb Start here. This tutorial describes some of the most basic functionality of the package and is the fastest way to get up-and-running with Monte-Carlo-based inference.
If you'd like to delve a little deeper into the model specification, there is also:
model-development.ipynb describes the (two-peak) model and how it is implemented.
model-testing.ipynb shows how to use the package to calculate the likelihood (which you almost always just want to do automatically with the emcee sampler as in the tutorial). It also does some very simple inference and tests (similar to the tests performed automatically by nosetests) to make sure the likelihood returns reasonable values.
dataprocessing.py includes a specification for the Spectrum object and other simple data processing tasks
io.py supports simple io for the Spectrum object
model.py specifies the likelihood for the model described above, and can evaluate them for Spectrum objects.
inferMC.py includes functions to perform Monte-Carlo sampling on the likelihood.
exampledata includes some light-weight example data, including simple data used for tests
tests contains unit tests
There are a few planned improvements.
-
Inclusion of various options to specify the priors on each parameter. Currently, the package only supports sampling over the likelihood and is therefore implicitly choosing improper uniform priors if the output is interpreted as a posterior.
-
MC with parallel tempering to improve convergence and reduce the need for fine-tuning of the guesses before fitting. (This second point has been somewhat obviated by starting the guesses at the maximum-likelihood peak.)
The package is released under the GNU GPLv3. See the LICENSE file.