vikram-s-narayan

# The purpose of this gist is to help pinpoint a potential bug in 
# QuasiMonteCarlo. This gist shows the working of the GEKPLS surrogate model 
# without the potential interferences from other Surrogates dependencies. 
# To run the gist and reproduce the results:
# Run this gist with QuasiMonteCarlo@0.2.19; it can be seen that it produces
# far better results without the first data point in the test dataset
# Additional notes: 
# - Ensure that no other packages that depend on QuasiMonteCarlo are in your package environment
# - QuasiMonteCarlo@0.2.17 and QuasiMonteCarlo@0.2.18 throw errors. More info for the cause can 
#   be seen here - https://github.com/SciML/QuasiMonteCarlo.jl/releases/tag/v0.2.17

vikram-s-narayan

# The purpose of this gist is to help pinpoint a potential bug in 
# QuasiMonteCarlo. This gist shows the working of the GEKPLS surrogate model 
# without the potential interferences from other Surrogates dependencies. 
# To run the gist and reproduce the results:
# Run this gist with QuasiMonteCarlo@0.2.16 => all tests pass (evidenced by 3 true prints)
# Then run this gist with QuasiMonteCarlo@0.2.19 => all tests fail in a fashion similar to 
# the failing tests of Surogates.jl - https://github.com/SciML/Surrogates.jl/pull/420
# Additional notes: 
# - Ensure that no other packages that depend on QuasiMonteCarlo are in your package environment
# - QuasiMonteCarlo@0.2.17 and QuasiMonteCarlo@0.2.18 throw errors. More info for the cause can 

vikram-s-narayan

Project Details:

Org: SciML-NumFOCUS

Contributor: Vikramaditya Narayan

Mentors: Chris Rackauckas and Ranjan Anantharaman

What Work Was Done:

vikram-s-narayan

#Add this line below 
#above https://github.com/SMTorg/smt/blob/4a4df255b9259965439120091007f9852f41523e/smt/surrogate_models/krg_based.py#L1165
#to try different theta values
optimal_theta = np.array([0.0001, 0.0001]) 

optimal_rlf_value, optimal_par = self._reduced_likelihood_function(
    theta=optimal_theta
)

vikram-s-narayan

using Surrogates
using Zygote
using SurrogatesPolyChaos
using SurrogatesAbstractGPs
using AbstractGPs

function vector_of_tuples_to_matrix(v)
    #helper function to convert training data generated by surrogate sampling into a matrix suitable for GEKPLS
    num_rows = length(v)
    num_cols = length(first(v))

vikram-s-narayan

using Surrogates
using Zygote
using SurrogatesPolyChaos
using SurrogatesAbstractGPs
using AbstractGPs

function vector_of_tuples_to_matrix(v)
    #helper function to convert training data generated by surrogate sampling into a matrix suitable for GEKPLS
    num_rows = length(v)
    num_cols = length(first(v))

vikram-s-narayan

using Surrogates
using Zygote
using SurrogatesPolyChaos

function vector_of_tuples_to_matrix(v)
    #helper function to convert training data generated by surrogate sampling into a matrix suitable for GEKPLS
    num_rows = length(v)
    num_cols = length(first(v))
    K = zeros(num_rows, num_cols)
    for row in 1:num_rows

vikram-s-narayan

using Statistics
using ScikitLearn
using LinearAlgebra

@sk_import cross_decomposition: PLSRegression

####### START OF ALL FUNCTIONS ##########

function _ge_compute_pls(X, y, n_comp, grads, delta_x, xlimits, extra_points)
    """

vikram-s-narayan

using LinearAlgebra
using ExtendableSparse

Base.copy(t::Tuple) = t

mutable struct RadialBasis{F,Q,X,Y,L,U,C,S,D} <: AbstractSurrogate
    phi::F
    dim_poly::Q
    x::X
    y::Y

vikram-s-narayan

using Surrogates
using LinearAlgebra
using Flux
using Flux: @epochs
using Zygote
using PolyChaos
using Test
#using Zygote: @nograd
#=
#FORWARD