sf_wrap

sf_wrap lifts arbitrary object-oriented python code to probabilistic programs expressed in numpyro via automatic method overwriting. It defines function decorators that implement stochastic functions and class decorators that augment the definitions of object methods. It is simple, self-contained, and permissively-licensed, with all functionality in a single file that you could copy into your project.

Usage example

Here's a silly class:

class A:
    def f(self, x: float):
        return x

    def g(self, x: float):
        return self.f(x) ** 2.0

There are two ways to use sf_wrap functionality: as context managers with existing code; or as class decorators when classes are defined. The mechanisms are equivalent for many purposes; which mechanism is used should be dictated by design of the larger software system.

Context managers

my_a = A()
x = 3.0
my_a_f_val = my_a.f(x)

with wrap.ClassTruncatedReal("f", low=-4, high=4, scale=0.1, cls=A,):
    my_stoch_a_f_val = my_a.f(x)

    with wrap.ClassTruncatedReal("g", low=-16, high=16, scale=0.1, cls=A):
        my_stoch_a_g_val = my_a.g(x)

my_a_f_val == 3.0, while my_stoch_a_f_val will deviate slightly from this value, as my_stoch_a_g_val will deviate slightly from my_stoch_a_f_val ** 2.

Class decorators

from sf_wrap import wrap


@wrap.ClassTruncatedReal("f", low=-3.0, high=3.0, scale=1.0,)
class RandomA(A): pass


@wrap.ClassTruncatedReal("g", low=-3.0, high=3.0, scale=1.0,)
@wrap.ClassTruncatedReal("f", low=-3.0, high=3.0, scale=1.0,)
class ReallyRandomA(A): pass

• Calling the f method of a RandomA object now returns the result of a truncated normal distribution with location parameter equal to the return value of the deterministic function A.f(self, x).
• Calling the g method of a ReallyRandomA object now returns the result of a truncated normal distribution with location parameter equal to the return value of the deterministic function A.g(self, y) where y is...
  • ...the result of a truncated normal distribution with location parameter equal to the the return value of the deterministic function A.f(self, x).

And so on and so forth. For more, please see the documentation.

Installation and testing

Installation is standard:

conda create --name my_env_name python=3.9
conda activate my_env_name
/stuff/my_env_name/bin/python -m pip install -r requirements.txt
/stuff/my_env_name/bin/python -m pip install .

Testing is standard:

/stuff/my_env_name/bin/python -m pytest --cov=sf_wrap

sf_wrap has been installed and tested on only MacOS (Catalina) and Ubuntu. It should work on any posix-compliant system, but YMMV. It has not been installed or tested on Windows.

Other information

sf_wrap is licensed under version 3 of the GNU Lesser General Public License (LGPL v3). Copyright Charles River Analytics, Inc., 2022 - present.

This material is based upon work supported by the Naval Information Warfare Center (NIWC) Atlantic under Contract No. N653622C8011. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the NIWC Atlantic. Approved for public release, distribution is unlimited.