Releases: usnistgov/optbayesexpt
Version 1.1.1
Version 1.1.1
May 27, 2021
-
Implemented just-in-time (jit) compilation of some of the most time-consuming methods using the
numba
package. Execution time
was shortened by 20 % to 40 %. A new demo program was added to highlight these capabilities.Since
numba
is not (yet) a required package foroptbayesexpt
, access tonumba
is tested and a BooleanGOT_NUMBA
is defined with scope extending over the whole optbayesexpt package.
May 21, 2021
-
Support for multi-channel measurements has been added to the OptBayesExpt class. As a result,
demos/lockin/obe_lockin.py
is no longer needed, and it has been removed. -
Support for noise parameter estimation is provided by a new component of the optbayesespt package,
OptBayesExptNoiseParam
, which takes anoise_parameter_index=(int)
argument to identify a parameter as measurement noise. These demos now useOptBayesExptNoiseParam
.demos/line_plus_noise/line_plus_noise.py
,demos/lockin/lockin_of_coil.py
, anddemos/sweeper/sweeper.py
-
Added support for
**kwargs
arguments to OptBayesExpt. Attribute values for OptBayesExpt, parent class ParticlePDF and OptBayesExpt child classes can now be set at instantiation. Keyword argumentsa_param
,resample_threshold
,auto_resample
andscale
are passed to ParticlePdf to tune resampling behavior.OptBayesExpt
useschoke
, andOptBayesExptNoiseParam
usesnoise_parameter_index
.
optbayesexpt v1.0.1
v1.0.1 is a stable version
April 27, 2020
Version 1.0.0 represents an overhaul of the optbayesexpt python package. It is not compatible with earlier versions, but only minor changes are needed to adapt script to use the new version. The most significant changes are briefly described here. Please consult the documentation at https://pages.nist.gov/optbayesexpt for more detail.
Probability Distribution Function:
Starting with V.1.0.0, the probability distribution function over parameter values is implemented using a sequential Monte Carlo scheme in ParticlePDF(), replacing the N-dimensional array representation used in ProbDistFunc(). This change boosts speed and allows more parameters in the model function.
Experiment Model:
Starting with V.1.0.0, the ExptModel class is no longer used. Methods of the ExptModel class are incorporated into OptBayesExpt.
OptBayesExpt class:
The OptBayesExpt class has been rewritten with reuse in mind. As much as possible, the calculations have been split out into separate methods. The goal was to make is easier to determine how to create customized child classes for different applications.
Creation of a functioning OptBayesExpt object has been simplified by including the model function, settings, parameters and constants as arguments to init(). In earlier versions, the object was created and then configured in separate steps.
Server:
The OBE_Server class has been redesigned to be a caretaker and TCP communication interface for OptBayesExpt objects. With this new design a OBE_Server object can initialize a series of OptBayesExpt objects with different configurations, e.g. for a series of measurement runs.
Initial release
Version 0.1.9 contains usable, mostly debugged code for sequential Bayesian experimental design.
Upcoming version will include a more efficient representation of probability density and abstracted modules for greater flexibility.