Computation time

Python usage
1

Hi Everyone,

I am new to OpenTURNS and I am wondering if it is possible to optimize this code ?
The objective is to generate a sample over multiple variables.

Thanks in advance

import numpy as np
import pandas as pd
import openturns as ot

distance_dist = ot.Uniform(300, 6000)
verticaldist  = ot.Uniform(1.6, 4)
epsilon = 0.01
nsamples=1000
fbottomFunction  = ot.SymbolicFunction(
    ["x0", "x1"], ["y0", "y1"], 
    f"y0 := max(min(((305+x0)*tan(x1*pi_/180)) + 152, 1990), 200) - {epsilon}; \
    y1 := 1990")

bottomDistribution = ot.BayesDistribution(
    ot.Uniform(), 
    ot.ComposedDistribution([distance_dist, verticaldist]), 
        fbottomFunction)

ftopFunction = ot.SymbolicFunction(["x0", "x1", "x2"], [f"x0+10", f"2000"])  
cDistribution= ot.BayesDistribution(ot.Uniform(), bottomDistribution, ftopFunction) 
cDistribution.setDescription(['top', 'base', 'dst', 'vdeg'])

fr  = ot.SymbolicFunction(["xTop", "xBase", "xDst", "xvdeg"], ["y0", "y1"], "y0 := (-10*(xDst<=2500)) + (-15*(2500<xDst)*(xDst<=4500)) + (-30*(4500<xDst)); \
                                                                               y1 := (10*(xDst<=2500)) + (15*(2500<xDst)*(xDst<=4500)) + (30*(4500<xDst));")
rDistribution = ot.BayesDistribution(ot.Uniform(), cDistribution, fr)  
rDistribution.setDescription(['rdeg', 'top', 'base', 'dst', 'vdeg'])

fy  = ot.SymbolicFunction(
    ["xR", "xTop", "xBase", "xDst", "xvdeg"], 
    ["y0", "y1"], "y0 := (-18.5*(xDst<=2500)) + (-24*(2500<xDst)); \
    y1 := (18.5*(xDst<=2500)) + (24*(2500<xDst));")
yDistribution = ot.BayesDistribution(ot.Uniform(), rDistribution, fy)
yDistribution.setDescription(['ydeg', 'rdeg', 'top', 'base', 'dst', 'vdeg'])

sample = yDistribution.getSample(10*nsamples)
yDistributionFinal = ot.UserDefined(
    sample, [1./w[0] for w in np.array(yDistribution.computePDF(sample))]) 
yDistributionFinal.setDescription(yDistribution.getDescription())
2

Hi Adil,
I had to change

ftopFunction = ot.SymbolicFunction(["x0", "x1", "x2"], [f"x0+10", f"2000"])

into

ftopFunction = ot.SymbolicFunction(["x0", "x1", "x2"], [f"x0+10", f"7000"])

in order to get rid of an error saying that you try to build a Uniform distribution with a lower bound greater than its upper bound. You should double check the distribution of x0 here.

With this modification, I get:

[regis@localhost Adil]$ time python code_adil.py 
WRN - class BayesDistribution is deprecated in favor of JointByConditioningDistribution

real    0m5,099s
user    0m6,078s
sys     0m2,585s

on my (quite powerful) laptop using the current master of OpenTURNS. We worked quite a little bit optimizing things in BayesDistribution (btw it has been renamed JointByConditioningDistribution). You can install our weekly build to try it (see Openturns | Anaconda.org).

Cheers

Régis

3

Hi Regis,

I will try the last OpenTURS version.

Thanks
Adil

4

Hi Régis,

with the version 1.23, the computation is faster. However, It seems that the variables order in the BayesDistribution output has changed between the 1.21 and 1.23. is it or did i missed something? How to get something compatible both versions?

Thanks
Regards
Adil

5

hello,
indeed, the conditioned/conditioning components were swapped in 1.23 for more efficiency
it seems missing from the changelog
you could use getMarginal with indices to reorder the components depending on the version

j