<s>
A	O
surrogate	B-Algorithm
model	I-Algorithm
is	O
an	O
engineering	O
method	O
used	O
when	O
an	O
outcome	O
of	O
interest	O
cannot	O
be	O
easily	O
measured	O
or	O
computed	O
,	O
so	O
an	O
approximate	O
mathematical	O
model	O
of	O
the	O
outcome	O
is	O
used	O
instead	O
.	O
</s>
<s>
One	O
way	O
of	O
alleviating	O
this	O
burden	O
is	O
by	O
constructing	O
approximation	O
models	O
,	O
known	O
as	O
surrogate	B-Algorithm
models	I-Algorithm
,	O
metamodels	O
or	O
emulators	O
,	O
that	O
mimic	O
the	O
behavior	O
of	O
the	O
simulation	O
model	O
as	O
closely	O
as	O
possible	O
while	O
being	O
computationally	O
cheaper	O
to	O
evaluate	O
.	O
</s>
<s>
Surrogate	B-Algorithm
models	I-Algorithm
are	O
constructed	O
using	O
a	O
data-driven	O
,	O
bottom-up	O
approach	O
.	O
</s>
<s>
When	O
only	O
a	O
single	O
design	O
variable	O
is	O
involved	O
,	O
the	O
process	O
is	O
known	O
as	O
curve	B-Algorithm
fitting	I-Algorithm
.	O
</s>
<s>
Though	O
using	O
surrogate	B-Algorithm
models	I-Algorithm
in	O
lieu	O
of	O
experiments	O
and	O
simulations	O
in	O
engineering	O
design	O
is	O
more	O
common	O
,	O
surrogate	O
modeling	O
may	O
be	O
used	O
in	O
many	O
other	O
areas	O
of	O
science	O
where	O
there	O
are	O
expensive	O
experiments	O
and/or	O
function	O
evaluations	O
.	O
</s>
<s>
Various	O
design	O
of	O
experiments	O
(	O
DOE	O
)	O
techniques	O
cater	O
to	O
different	O
sources	O
of	O
errors	O
,	O
in	O
particular	O
,	O
errors	O
due	O
to	O
noise	O
in	O
the	O
data	O
or	O
errors	O
due	O
to	O
an	O
improper	O
surrogate	B-Algorithm
model	I-Algorithm
.	O
</s>
<s>
Popular	O
surrogate	O
modeling	O
approaches	O
are	O
:	O
polynomial	O
response	O
surfaces	O
;	O
kriging	O
;	O
more	O
generalized	O
Bayesian	O
approaches	O
,	O
gradient-enhanced	O
kriging	O
(	O
GEK	O
)	O
;	O
radial	B-Algorithm
basis	I-Algorithm
function	I-Algorithm
;	O
support	B-Algorithm
vector	I-Algorithm
machines	I-Algorithm
;	O
space	B-Algorithm
mapping	I-Algorithm
;	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
and	O
Bayesian	O
networks	O
.	O
</s>
<s>
For	O
some	O
problems	O
,	O
the	O
nature	O
of	O
true	O
function	O
is	O
not	O
known	O
a	O
priori	O
so	O
it	O
is	O
not	O
clear	O
which	O
surrogate	B-Algorithm
model	I-Algorithm
will	O
be	O
most	O
accurate	O
.	O
</s>
<s>
In	O
these	O
cases	O
,	O
physics-based	O
surrogates	O
such	O
as	O
space-mapping	B-Algorithm
based	O
models	O
are	O
the	O
most	O
efficient	O
.	O
</s>
<s>
Spanning	O
two	O
decades	O
of	O
development	O
and	O
engineering	O
applications	O
,	O
Rayas-Sanchez	O
reviews	O
aggressive	O
space	B-Algorithm
mapping	I-Algorithm
exploiting	O
surrogate	B-Algorithm
models	I-Algorithm
.	O
</s>
<s>
have	O
published	O
a	O
state-of-the-art	O
review	O
of	O
surrogate	B-Algorithm
models	I-Algorithm
used	O
in	O
water	O
resources	O
management	O
field	O
.	O
</s>
<s>
Recently	O
proposed	O
comparison-based	O
surrogate	B-Algorithm
models	I-Algorithm
(	O
e.g.	O
</s>
<s>
ranking	O
support	B-Algorithm
vector	I-Algorithm
machine	I-Algorithm
)	O
for	O
evolutionary	B-Algorithm
algorithms	I-Algorithm
,	O
such	O
as	O
CMA-ES	B-Algorithm
,	O
allow	O
to	O
preserve	O
some	O
invariance	O
properties	O
of	O
surrogate-assisted	O
optimizers	O
:	O
</s>
<s>
An	O
important	O
distinction	O
can	O
be	O
made	O
between	O
two	O
different	O
applications	O
of	O
surrogate	B-Algorithm
models	I-Algorithm
:	O
design	O
optimization	O
and	O
design	O
space	O
approximation	O
(	O
also	O
known	O
as	O
emulation	O
)	O
.	O
</s>
<s>
In	O
surrogate	B-Algorithm
model	I-Algorithm
based	O
optimization	O
,	O
an	O
initial	O
surrogate	O
is	O
constructed	O
using	O
some	O
of	O
the	O
available	O
budgets	O
of	O
expensive	O
experiments	O
and/or	O
simulations	O
.	O
</s>
<s>
The	O
remaining	O
experiments/simulations	O
are	O
run	O
for	O
designs	O
which	O
the	O
surrogate	B-Algorithm
model	I-Algorithm
predicts	O
may	O
have	O
promising	O
performance	O
.	O
</s>
<s>
Search	O
surrogate	B-Algorithm
model	I-Algorithm
(	O
the	O
model	O
can	O
be	O
searched	O
extensively	O
,	O
e.g.	O
</s>
<s>
This	O
package	O
provides	O
a	O
library	O
of	O
surrogate	B-Algorithm
models	I-Algorithm
that	O
is	O
simple	O
to	O
use	O
and	O
facilitates	O
the	O
implementation	O
of	O
additional	O
methods	O
.	O
</s>
<s>
It	O
also	O
includes	O
new	O
surrogate	B-Algorithm
models	I-Algorithm
that	O
are	O
not	O
available	O
elsewhere	O
:	O
kriging	O
by	O
partial-least	O
squares	O
reduction	O
and	O
energy-minimizing	O
spline	O
interpolation	O
.	O
</s>
<s>
:	O
is	O
a	O
Julia	B-Application
packages	O
which	O
offers	O
tools	O
like	O
random	B-Algorithm
forests	I-Algorithm
,	O
radial	O
basis	O
methods	O
and	O
kriging	O
.	O
</s>
