I have a decision variable declaration that does not work in Pyomo 5.7 that did work in 5.6.9.
The line in question is part of building a (concrete) model:
model.v = pe.Var(within = pe.RealSet) #pe = pyomo.environ
This ran fine when I had Pyomo version 5.6.9, but now gives errors described below in the provided code block.
To run with 5.7 I had to drop the within statement so I am left with:
model.v = pe.Var()
What is the reasoning for no longer supporting an explicit statement that the domain includes all reals?
When running with the original statement with Pyomo 5.7, I get this error (CPython 3.7.1 on Windows 10):
ERROR: Constructing component 'v' from data=None failed: TypeError: __new__()
takes 1 positional argument but 2 were given
Traceback (most recent call last):
File "runscript.py", line 216, in <module>
payoff_matrix = my_matrix
File "C:\Users\myusername\Documents\modelbuildscript.py", line 24, in __init__
model.v = pe.Var(within = pe.RealSet)
File "C:\EngTools\Anaconda3\2018.12\lib\site-packages\pyomo\core\base\block.py", line 543, in __setattr__
self.add_component(name, val)
File "C:\EngTools\Anaconda3\2018.12\lib\site-packages\pyomo\core\base\block.py", line 1081, in add_component
val.construct(data)
File "C:\EngTools\Anaconda3\2018.12\lib\site-packages\pyomo\core\base\var.py", line 613, in construct
self._initialize_members((None,))
File "C:\EngTools\Anaconda3\2018.12\lib\site-packages\pyomo\core\base\var.py", line 682, in_initialize_members
self.domain = self._domain_init_rule(self._parent())
TypeError: __new__() takes 1 positional argument but 2 were given
The short answer is that you should use
model.v = pe.Var(within=pe.Reals)
The slightly longer answer is that the Pyomo global sets (Reals, NonNegativeReals, Integers, PositiveIntegers, Binary, etc.) were originally defined using their own class hierarchy independent of the Set and RangeSet components. Unfortunately, that led to numerous problems (correctly detecting discrete sets, performing set operations using global sets, etc). Pyomo 5.7 switched to a new rewrite of the Set / RangeSet components that among numerous other bug fixes moved to a new paradigm where all sets in Pyomo were either Set or RangeSet components (even global sets). We put in a backwards compatibility layer that caught all the legacy uses of the RealSet class in the test suite, but your use case (while it worked) was not a documented / tested case. PR 1619 will restore this functionality (likely will be included in Pyomo 5.7.1), and adds an official deprecation warning for use of RealSet.