SRDS is mainly a wrapper around scipy's statistical functions
(scipy.stats).
It makes it easier to sample from parameterized distributions or populations by providing a uniform interface, the
RandomSampler
.
It provides several convenience methods and classes to improve code that uses statistical distribution sampling, e.g.,
in simulators.
srds adds several classes that make it easier to utilize scipy statistical distributions.
To sample from a log-normal distribution with σ = 0.25
, but truncate at 0.5
and 2
, the BoundRejectionSampler
helps:
from srds import ParameterizedDistribution as PDist, BoundRejectionSampler
dist = PDist.lognorm(0.25)
sampler = BoundRejectionSampler(dist, 0.5, 2)
x = sampler.sample(10)
calling dist.rvs
on a scipy statistical distribution is computationally expensive. This is problematic for code that
does something like:
# will be slow (calls dist.rvs 10000 times)
for i in range(10000):
x = dist.sample()
# ...
srds provides a BufferedSampler
that draws a larger sample from a distribution, and subsequently returns from that
sample.
from srds import BufferedSampler
dist = BufferedSampler(dist)
# will be much faster! (calls dist.rvs only 10 times with a sample size of 1k)
for i in range(10000):
x = dist.sample()
# ...
You can use the PopulationSampler
to draw from a discrete set, and also bias the sampling with weights.
from srds import PopulationSampler
sampler = PopulationSampler(['a', 'b', 'c'], [8, 1, 1])
sampler.sample() # will return 'a' 8 out of 10 times on average
sampler.sample(10) # returns a list containing items from ['a', 'b', 'c'] in random order