Create a multivariate basis of functions from scalar multivariable functions¶
We want to build a finite basis of functions where each with , from given scalar multivariable functions .
In other words, we want to create a basis of functions having both a multivariate input and output from functions having a multivariate input and a scalar output.
In this example, we do that by the aggregation of the given functions ,
AggregatedFunction class. This class allows one to create a multivariate output
function from a list of functions having a univariate output dimension. Note that we can only aggregate functions
that share the same input space.
We illustrate two cases that differ from the way the functions have been created:
Case 1: we suppose that the functions have no particular origin,
Case 2: we suppose that the functions have been built from particular bases.
Case 1: Aggregation of given functions¶
In that case, we have some functions and we aggregate them in order to build the basis .
For example, we set , and we use the functions defined by:
and we want to build the finite basis defined by:
import openturns as ot d = 2 q = 3 # Create the functions g_i g_1 = ot.SymbolicFunction(['x1', 'x2'], ['1']) g_2 = ot.SymbolicFunction(['x1', 'x2'], ['x1+x2']) g_3 = ot.SymbolicFunction(['x1', 'x2'], ['x1^2+x2^2']) # Create the multivariate basis f_1 = ot.AggregatedFunction([g_1] * q) f_2 = ot.AggregatedFunction([g_1, g_2, g_3]) basis = ot.Basis([f_1, f_2]) print('Fonction f_1 : ', basis.build(0))
Fonction f_1 : [[x1,x2]->,[x1,x2]->,[x1,x2]->]
Case 2: Aggregation of functions built from given bases¶
In that case, we use some functions that have been built from particular bases. For example, they are members of polynomial families which are orthonormal to given distributions.
In this example, we use both polynomial families:
which is a orthonormal multivariate polynomial basis with respect to the uniform distributipn on (it is obtained through a tensor product of univariate Legendre polynomials orthonormal with respect to the uniform distribution on ),
which is a orthonormal multivariate polynomial basis with respect to the multivariate standard normal distribution (it is obtained through a tensor product of univariate Hermite polynomials orthonormal with respect to the standard normal univariate distribution).
We consider the linear enumeration of the polynomial family. The functions are built as the aggregation of the -th polynomials of the families and . We only consider and of total degree less than 2: then we have 6 polynomials .
We still set , .
d = 2 legendre_builder = ot.OrthogonalProductPolynomialFactory([ot.LegendreFactory()] * d) hermite_builder = ot.OrthogonalProductPolynomialFactory([ot.HermiteFactory()] * d) # Generate the multivariate functions f_i (quadratic) total_deg = 2 basis_size = ot.LinearEnumerateFunction(d).getStrataCumulatedCardinal(total_deg) basis =  for i in range(basis_size): f_i = ot.AggregatedFunction([legendre_builder.build(i), hermite_builder.build(i)]) basis.append(f_i) basis = ot.Basis(basis)