LHS-MDU

Installation

You can install lhsmdu using pip:

$ pip install lhsmdu

Alternatively, you can clone on github and then install the package locally:

$ git clone https://github.com/sahilm89/lhsmdu
$ cd lhsmdu
$ python setup.py install --user   # for this user only.

or:

$ pip install git+https://github.com/sahilm89/lhsmdu --user

Basics

This is a package for generating latin hypercube samples with multi-dimensional uniformity.

To use, simply do:

>>> import lhsmdu
>>> k = lhsmdu.sample(2, 20) # Latin Hypercube Sampling with multi-dimensional uniformity

This will generate a nested list with 2 variables, with 20 samples each.

To plot and see the difference between Monte Carlo and LHS-MDU sampling for a 2 dimensional system:

>>> l = lhsmdu.createRandomStandardUniformMatrix(2, 20) # Monte Carlo sampling
>>> k = lhsmdu.sample(2, 20) # Latin Hypercube Sampling with multi-dimensional uniformity
>>> k = np.array(k)
>>> l = np.array(l)
>>> import matplotlib.pyplot as plt
>>> fig = plt.figure()
>>> ax = fig.gca()
>>> ax.set_xticks(numpy.arange(0,1,0.1))
>>> ax.set_yticks(numpy.arange(0,1,0.1))
>>> plt.scatter(k[0], k[1], color="g", label="LHS-MDU")
>>> plt.scatter(l[0], l[1], color="r", label="MC")
>>> plt.grid()
>>> plt.show()

You can use the strata generated by the algorithm to sample again, if you so desire. For this, you can do:

>>> m = lhsmdu.resample()
>>> n = lhsmdu.resample()
>>> o = lhsmdu.resample()

This will again generate the same number of samples as before, a nested list with 2 variables, with 20 samples each.

You can plot these together and see the sampling from the strata::

>>> m = np.array(m)
>>> n = np.array(n)
>>> o = np.array(o)

>>> fig = plt.figure()
>>> ax = fig.gca()
>>> ax.set_xticks(numpy.arange(0,1,0.1))
>>> ax.set_yticks(numpy.arange(0,1,0.1))
>>> plt.title("LHS-MDU")
>>> plt.scatter(k[0], k[1], c="g", label="sample 1")
>>> plt.scatter(m[0], m[1], c="r", label="resample 2")
>>> plt.scatter(n[0], n[1], c="b", label="resample 3")
>>> plt.scatter(o[0], o[1], c="y", label="resample 4")
>>> plt.grid()
>>> plt.show()

Alternatively, you can choose to get new strata each time, and see the sampling hence:

>>> p = lhsmdu.sample(2, 20) # Latin Hypercube Sampling with multi-dimensional uniformity
>>> q = lhsmdu.sample(2, 20) # Latin Hypercube Sampling with multi-dimensional uniformity
>>> r = lhsmdu.sample(2, 20) # Latin Hypercube Sampling with multi-dimensional uniformity

>>> p = np.array(p)
>>> q = np.array(q)
>>> r = np.array(r)

>>> fig = plt.figure()
>>> ax = fig.gca()
>>> ax.set_xticks(numpy.arange(0,1,0.1))
>>> ax.set_yticks(numpy.arange(0,1,0.1))
>>> plt.title("LHS-MDU")
>>> plt.scatter(k[0], k[1], c="g", label="sample 1")
>>> plt.scatter(p[0], p[1], c="r", label="sample 2")
>>> plt.scatter(q[0], q[1], c="b", label="sample 3")
>>> plt.scatter(r[0], r[1], c="y", label="sample 4")
>>> plt.grid()
>>> plt.show()

Changing the random seed

You will notice that the strata generated are the same each time you run the program again. This is because the random seed is a global constant set to a default value by design, so that simulations can be replicated. In order to change this behavior, you can set a new random seed using the following code:

>>> randSeed = 11 # random number of choice
>>> lhsmdu.setRandomSeed(randSeed) # Latin Hypercube Sampling with multi-dimensional uniformity
>>> lhsmdu.sample(2, 20) # Latin Hypercube Sampling with multi-dimensional uniformity

Alternatively, you can also set the seed by using sample with a new seed:

>>> lhsmdu.sample(2, 20, randomSeed=randSeed) # Latin Hypercube Sampling with multi-dimensional uniformity

To change the random seed in every run, you can set on top of the program:

>>> lhsmdu.setRandomSeed(None)

Sampling from arbitrary CDFs

After uniformly distributed samples have been generated from LHSMDU, you can convert these to samples from arbitrary distributions using inverse tranform sampling. In this, the CDF [0,1] of the distribution of interest is inverted, and then data points corresponding to the uniformly sampled points are picked up. To do this, you must have a rv_contiuous or rv_discrete distribution instance taken from scipy.stats. You can also use frozen distributions (after setting loc and scale parameters). Following is an example for normal distribution.:

>>> import scipy.stats.distributions as ssd
>>> p = ssd.norm
>>> new_samples = lhsmdu.inverseTransformSample(p, k[0])
>>> plt.hist(new_samples[0])
>>> plt.show()

Citing this repository

To cite, please cite both the original paper from Deutsch and Deutsch: http://dx.doi.org/10.1016%2Fj.jspi.2011.09.016. and the repository: https://doi.org/10.5281/zenodo.2578781