toledo60 / pylhd Goto Github PK

Hi there,

Nice package you got there!

Could you provide an example of how to use a design to produce random numbers from the desired distribution?

Say that I would like a 2D random Latin hypercube with $x \in [-2, 0.5)$ and $y \in [-2, 2)$.

Is it possible to create that with pyLHD?

Hi,
thanks for your nice package!
I would like to include it in this list of mine https://github.com/danieleongari/awesome-design-of-experiments.

However, first we need to investigate an unexpected behavior.
I may use your package to select a "MaxPro-LHD", and therefore I need to optimize the design with SA_LHD or LA_LHD with criterion='MaxProCriterion'. The problem is that both functions underperform a simple random search like:

mpc_min = 1e100
X_best = None
for i in range(10000):
    Xi = pl.rLHD(nrows=10,ncols=6)
    mpc = pl.MaxProCriterion(Xi)
    if mpc < mpc_min:
        mpc_min = mpc
        X_best = Xi

print("MaxPro Criterion:", pl.MaxProCriterion(X_best))
X_best

which runs in a comparable amount of time (under 3 seconds).
Also increasing the settings for a finer optimization in the two functions (more iterations) the situation does not solve.

Any idea?