1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
|
\section{Discussion}
\label{chap:disc}
A major driver for the development of the \pcasp is the Pyomo optimization software~\cite{Pyomo}.
Pyomo defines a variety of component interfaces that can be used to
customized and extend Pyomo's management of optimization solvers. This includes interfaces for components that
\begin{itemize}
\item write files that define an optimization problem
\item convert files into a format that is compatible with an optimizer
\item execute an optimizer
\item read files that define optimizer results and execution status
\end{itemize}
The \pcasp has had a fundamental impact on the design of Pyomo because it
supports a new software design for optimization frameworks.
The typical object oriented approach for optimization software is
to use classes and class inheritance. For example, the OPT++~\cite{Mez94}
optimization software library defines base classes with different
characteristics (e.g. differentiability), and a concrete optimization
solver is instantiated as a subclass of an appropriate base class. In this
context, the base class can be viewed as defining the interface for the
solvers that inherit from it.
Pyomo components leverage the \pcasp to separate the declaration
of component interfaces from their implementation. For example,
the interface to optimization solvers are again declared with a class.
However, solver plugins are not required to be subclasses of the interface
class. Instead, they are simply required to provide the same interface
methods. Consequently, Pyomo can be extended and configured in a modular
manner that is qualitatively different from other optimization frameworks.
The PCA allows Pyomo to dynamically construct optimization strategies and
combine independently-developed modeling, reformulation, preprocessing,
and optimization approaches in a manner that is substantially more
flexible and extensible compared to other widely used optimization
frameworks.
|
