Gaussian Floats - imagining a programming language that represents the uncertainties of the real-world.

gaussian_float.py

from scipy import stats

"""
Imagining a programming language where basic types are represented as a Gaussian distribution. Would make for some pretty interesting (read: hair-pulling) programming experience, lol.
. The default assumptions in Deep Learning these days. But in a way, this may be how we should model the world around us. There are no statics or constants, everything is a probability distribution function.
"""

class GaussianFloat(object):

    def __init__(self, val, scale=None):
        self._val = float(val)
        self._scale = scale or 1.0
        self._dist = stats.norm(self._val, self._scale)

    def get(self):
        return self._dist.rvs(size=1)[0]

    def __eq__(self, obj):
        return type(obj) == type(self) and obj._val == self._val and obj._scale == self._scale

    def __add__(self, obj):
        assert type(self) == type(obj)
        new_val = self._val + obj._val
        new_scale = self._scale + obj._scale  # This is prob wrong math
        return self.__class__(new_val, new_scale)

    def __mul__(self, obj):
        assert type(self) == type(obj)
        new_val = self._val * obj._val
        new_scale = self._scale * obj._scale  # This is prob wrong math
        return self.__class__(new_val, new_scale)