Export a metamodel

In this example we will see how to export a metamodel from the context it is created to another context where it will actually be used with the help of the pickle module.

import openturns as ot
from openturns.usecases import cantilever_beam
import pickle

Load the cantilever beam use-case. We want to use the independent distribution to get meaningful Sobol’ indices.

beam = cantilever_beam.CantileverBeam()
g = beam.model
distribution = beam.independentDistribution

Generate a learning sample with Monte-Carlo simulation (or retrieve the design from experimental data).

ot.RandomGenerator.SetSeed(0)
N = 30  # size of the experimental design
X = distribution.getSample(N)
Y = g(X)

Build a chaos metamodel

algo = ot.FunctionalChaosAlgorithm(X, Y, distribution)
algo.run()
assert algo.getResult().getResiduals()[0] < 1e-12
metamodel = algo.getResult().getMetaModel()

Save the metamodel into a .pkl file for later use

with open("metamodel.pkl", "wb") as f:
    pickle.dump(metamodel, f)

Reload the metamodel in another context from the same .pkl

with open("metamodel.pkl", "rb") as f:
    metamodel = pickle.load(f)

Reuse the loaded metamodel

x = [6.70455e10, 300.0, 2.55, 1.45385e-07]
y = metamodel(x)
print(y)

[0.170031]