A better scipy.stats.uniform
The stats sub-package of scipy is quite cool.
In particular, it provides dozens of probability distributions implemented with
a common interface.
But scipy.stats.uniform always bugged me.
>>> from scipy import stats
>>> help(stats.uniform)
A uniform continuous random variable.
This distribution is constant between `loc` and ``loc + scale``.Why loc + scale? Why not scale?
So I wrote better_uniform: eight small lines of code that don't bug me so
much.
from scipy import stats
class frozen(stats._distn_infrastructure.rv_continuous_frozen):
def __init__(self, dist, *args, **kwds):
super(frozen,self).__init__(dist, *args, **kwds)
def buniform(a, b): # b for better
dist = stats.uniform
dist.name = 'uniform'
return frozen(dist, loc=a, scale=b-a)Now it works as I expect it to work:
d = buniform(0, 1)
d.rvs() # 0 < rv < 1
d.suport() # (0.0, 1.0)
d = buniform(1, 2)
d.rvs() # 1 < rv < 2
d.support() # (1.0, 2.0)
# note the difference
from scipy.stats import uniform
d = uniform(1, 2)
d.rvs() # 1 < rv < 3
d.support() # (1.0, 3.0)That's it!
Cool, I want it!
pip install better-uniform
or
git clone https://github.com/j-faria/better_uniform.git
cd better_uniform
python setup.py installand later, from Python
from better_uniform import buniformor better yet
from better_uniform import buniform as uniform
