class ConvexHull_Transformation(Transformation):
"""Relax disjunctive model by forming the convex hull.
Relaxes a disjunctive model into an algebraic model by forming the
convex hull of each disjunction.
This transformation accepts the following keyword arguments:
Parameters
----------
perspective_function : str
The perspective function used for the disaggregated variables.
Must be one of 'FurmanSawayaGrossmann' (default),
'LeeGrossmann', or 'GrossmannLee'
EPS : float
The value to use for epsilon [default: 1e-4]
targets : (block, disjunction, ComponentUID or list of those types)
The targets to transform. This can be a block, disjunction, or a
list of blocks and Disjunctions [default: the instance]
After transformation, every transformed disjunct will have a
"_gdp_transformation_info" dict containing 2 entries:
'relaxed': True,
'chull': {
'relaxationBlock': <block>,
'relaxedConstraints': ComponentMap(constraint: relaxed_constraint)
'disaggregatedVars': ComponentMap(var: list of disaggregated vars),
'bigmConstraints': ComponentMap(disaggregated var: bigM constraint),
}
In addition, any block or disjunct containing a relaxed disjunction
will have a "_gdp_transformation_info" dict with the following
entry:
'disjunction_or_constraint': <constraint>
Finally, the transformation will create a new Block with a unique
name beginning "_pyomo_gdp_chull_relaxation". That Block will
contain an indexed Block named "relaxedDisjuncts", which will hold
the relaxed disjuncts. This block is indexed by an integer
indicating the order in which the disjuncts were relaxed. Each
block will have a "_gdp_transformation_info" dict with the following
entries:
'src': <source disjunct>
'srcVars': ComponentMap(disaggregated var: original var),
'srcConstraints': ComponentMap(relaxed_constraint: constraint)
'boundConstraintToSrcVar': ComponentMap(bigm_constraint: orig_var),
"""
alias('gdp.chull', doc=textwrap.fill(textwrap.dedent(__doc__.strip())))
CONFIG = cfg.ConfigBlock('gdp.chull')
CONFIG.declare(
'targets',
cfg.ConfigValue(
default=None,
domain=target_list,
description="target or list of targets that will be relaxed",
doc="""
This specifies the target or list of targets to relax as either a
component, ComponentUID, or string that can be passed to a
ComponentUID; or an iterable of these types. If None (default),
the entire model is transformed."""))
CONFIG.declare(
'perspective function',
cfg.ConfigValue(
default='FurmanSawayaGrossmann',
domain=cfg.In(
['FurmanSawayaGrossmann', 'LeeGrossmann', 'GrossmannLee']),
description='perspective function used for variable disaggregation',
doc="""
The perspective function used for variable disaggregation
"LeeGrossmann" is the original NL convex hull from Lee &
Grossmann (2000) [1]_, which substitutes nonlinear constraints
h_ik(x) <= 0
with
x_k = sum( nu_ik )
y_ik * h_ik( nu_ik/y_ik ) <= 0
"GrossmannLee" is an updated formulation from Grossmann &
Lee (2003) [2]_, which avoids divide-by-0 errors by using:
x_k = sum( nu_ik )
(y_ik + eps) * h_ik( nu_ik/(y_ik + eps) ) <= 0
"FurmanSawayaGrossmann" (default) is an improved relaxation [3]_
that is exact at 0 and 1 while avoiding numerical issues from
the Lee & Grossmann formulation by using:
x_k = sum( nu_ik )
((1-eps)*y_ik + eps) * h_ik( nu_ik/((1-eps)*y_ik + eps) ) \
- eps * h_ki(0) * ( 1-y_ik ) <= 0
References
----------
.. [1] Lee, S., & Grossmann, I. E. (2000). New algorithms for
nonlinear generalized disjunctive programming. Computers and
Chemical Engineering, 24, 2125-2141
.. [2] Grossmann, I. E., & Lee, S. (2003). Generalized disjunctive
programming: Nonlinear convex hull relaxation and algorithms.
Computational Optimization and Applications, 26, 83-100.
.. [3] Furman, K., Sawaya, N., and Grossmann, I. A computationally
useful algebraic representation of nonlinear disjunctive convex
sets using the perspective function. Optimization Online
(2016). http://www.optimization-online.org/DB_HTML/2016/07/5544.html.
"""))
CONFIG.declare(
'EPS',
cfg.ConfigValue(
default=1e-4,
domain=cfg.PositiveFloat,
description="Epsilon value to use in perspective function",
))
def __init__(self):
super(ConvexHull_Transformation, self).__init__()
self.handlers = {
Constraint: self._xform_constraint,
Var: False,
Connector: False,
Expression: False,
Param: False,
Set: False,
Suffix: False,
Disjunction: self._warn_for_active_disjunction,
Disjunct: self._warn_for_active_disjunct,
Block: self._transform_block_on_disjunct,
}
def _apply_to(self, instance, **kwds):
self._config = self.CONFIG(kwds.pop('options', {}))
self._config.set_value(kwds)
# make a transformation block
transBlockName = unique_component_name(instance,
'_pyomo_gdp_chull_relaxation')
transBlock = Block()
instance.add_component(transBlockName, transBlock)
transBlock.relaxedDisjuncts = Block(Any)
transBlock.lbub = Set(initialize=['lb', 'ub', 'eq'])
transBlock.disjContainers = ComponentSet()
targets = self._config.targets
if targets is None:
targets = (instance, )
_HACK_transform_whole_instance = True
else:
_HACK_transform_whole_instance = False
for _t in targets:
t = _t.find_component(instance)
if t is None:
raise GDP_Error(
"Target %s is not a component on the instance!" % _t)
if t.type() is Disjunction:
if t.parent_component() is t:
self._transformDisjunction(t, transBlock)
else:
self._transformDisjunctionData(t, transBlock, t.index())
elif t.type() in (Block, Disjunct):
if t.parent_component() is t:
self._transformBlock(t, transBlock)
else:
self._transformBlockData(t, transBlock)
else:
raise GDP_Error(
"Target %s was not a Block, Disjunct, or Disjunction. "
"It was of type %s and can't be transformed" %
(t.name, type(t)))
# Go through our dictionary of indexed things and deactivate
# the containers that don't have any active guys inside of
# them. So the invalid component logic will tell us if we
# missed something getting transformed.
for obj in transBlock.disjContainers:
if not obj.active:
continue
for i in obj:
if obj[i].active:
break
else:
# HACK due to active flag implementation.
#
# Ideally we would not have to do any of this (an
# ActiveIndexedComponent would get its active status by
# querring the active status of all the contained Data
# objects). As a fallback, we would like to call:
#
# obj._deactivate_without_fixing_indicator()
#
# However, the sreaightforward implementation of that
# method would have unintended side effects (fixing the
# contained _DisjunctData's indicator_vars!) due to our
# class hierarchy. Instead, we will directly call the
# relevant base class (safe-ish since we are verifying
# that all the contained _DisjunctionData are
# deactivated directly above).
ActiveComponent.deactivate(obj)
# HACK for backwards compatibility with the older GDP transformations
#
# Until the writers are updated to find variables on things
# other than active blocks, we need to reclassify the Disjuncts
# as Blocks after transformation so that the writer will pick up
# all the variables that it needs (in this case, indicator_vars).
if _HACK_transform_whole_instance:
HACK_GDP_Disjunct_Reclassifier().apply_to(instance)
def _contained_in(self, var, block):
"Return True if a var is in the subtree rooted at block"
while var is not None:
if var.parent_component() is block:
return True
var = var.parent_block()
if var is block:
return True
return False
def _transformBlock(self, obj, transBlock):
for i in sorted(iterkeys(obj)):
self._transformBlockData(obj[i], transBlock)
def _transformBlockData(self, obj, transBlock):
# Transform every (active) disjunction in the block
for disjunction in obj.component_objects(
Disjunction,
active=True,
sort=SortComponents.deterministic,
descend_into=(Block, Disjunct),
descent_order=TraversalStrategy.PostfixDFS):
self._transformDisjunction(disjunction, transBlock)
def _getDisjunctionConstraints(self, disjunction):
# Put the disjunction constraint on its parent block
# We never do this for just a DisjunctionData because we need
# to know about the index set of its parent component. So if
# we called this on a DisjunctionData, we did something wrong.
assert isinstance(disjunction, Disjunction)
parent = disjunction.parent_block()
if hasattr(parent, "_gdp_transformation_info"):
infodict = parent._gdp_transformation_info
if type(infodict) is not dict:
raise GDP_Error(
"Component %s contains an attribute named "
"_gdp_transformation_info. The transformation requires "
"that it can create this attribute!" % parent.name)
try:
# On the off-chance that another GDP transformation went
# first, the infodict may exist, but the specific map we
# want will not be present
orConstraintMap = infodict['disjunction_or_constraint']
except KeyError:
orConstraintMap = infodict['disjunction_or_constraint'] \
= ComponentMap()
try:
disaggregationConstraintMap = infodict[
'disjunction_disaggregation_constraints']
except KeyError:
disaggregationConstraintMap = infodict[
'disjunction_disaggregation_constraints'] \
= ComponentMap()
else:
infodict = parent._gdp_transformation_info = {}
orConstraintMap = infodict['disjunction_or_constraint'] \
= ComponentMap()
disaggregationConstraintMap = infodict[
'disjunction_disaggregation_constraints'] \
= ComponentMap()
if disjunction in disaggregationConstraintMap:
disaggregationConstraint = disaggregationConstraintMap[disjunction]
else:
# add the disaggregation constraint
disaggregationConstraint \
= disaggregationConstraintMap[disjunction] = Constraint(Any)
parent.add_component(
unique_component_name(parent, '_gdp_chull_relaxation_' + \
disjunction.name + '_disaggregation'),
disaggregationConstraint)
# If the Constraint already exists, return it
if disjunction in orConstraintMap:
orC = orConstraintMap[disjunction]
else:
# add the XOR (or OR) constraints to parent block (with
# unique name) It's indexed if this is an
# IndexedDisjunction, not otherwise
orC = Constraint(disjunction.index_set()) if \
disjunction.is_indexed() else Constraint()
parent.add_component(
unique_component_name(
parent,
'_gdp_chull_relaxation_' + disjunction.name + '_xor'), orC)
orConstraintMap[disjunction] = orC
return orC, disaggregationConstraint
def _transformDisjunction(self, obj, transBlock):
# create the disjunction constraint and disaggregation
# constraints and then relax each of the disjunctionDatas
for i in sorted(iterkeys(obj)):
self._transformDisjunctionData(obj[i], transBlock, i)
# deactivate so we know we relaxed
obj.deactivate()
def _transformDisjunctionData(self, obj, transBlock, index):
# Convex hull doesn't work if this is an or constraint. So if
# xor is false, give up
if not obj.xor:
raise GDP_Error(
"Cannot do convex hull transformation for "
"disjunction %s with or constraint. Must be an xor!" %
obj.name)
parent_component = obj.parent_component()
transBlock.disjContainers.add(parent_component)
orConstraint, disaggregationConstraint \
= self._getDisjunctionConstraints(parent_component)
# We first go through and collect all the variables that we
# are going to disaggregate.
varOrder_set = ComponentSet()
varOrder = []
varsByDisjunct = ComponentMap()
for disjunct in obj.disjuncts:
# This is crazy, but if the disjunct has been previously
# relaxed, the disjunct *could* be deactivated.
not_active = not disjunct.active
if not_active:
disjunct._activate_without_unfixing_indicator()
try:
disjunctVars = varsByDisjunct[disjunct] = ComponentSet()
for cons in disjunct.component_data_objects(
Constraint,
active=True,
sort=SortComponents.deterministic,
descend_into=Block):
# we aren't going to disaggregate fixed
# variables. This means there is trouble if they are
# unfixed later...
for var in EXPR.identify_variables(cons.body,
include_fixed=False):
# Note the use of a list so that we will
# eventually disaggregate the vars in a
# deterministic order (the order that we found
# them)
disjunctVars.add(var)
if var not in varOrder_set:
varOrder.append(var)
varOrder_set.add(var)
finally:
if not_active:
disjunct._deactivate_without_fixing_indicator()
# We will only disaggregate variables that
# 1) appear in multiple disjuncts, or
# 2) are not contained in this disjunct, or
# 3) are not themselves disaggregated variables
varSet = []
localVars = ComponentMap((d, []) for d in obj.disjuncts)
for var in varOrder:
disjuncts = [d for d in varsByDisjunct if var in varsByDisjunct[d]]
if len(disjuncts) > 1:
varSet.append(var)
elif self._contained_in(var, disjuncts[0]):
localVars[disjuncts[0]].append(var)
elif self._contained_in(var, transBlock):
# There is nothing to do here: these are already
# disaggregated vars that can/will be forced to 0 when
# their disjunct is not active.
pass
else:
varSet.append(var)
# Now that we know who we need to disaggregate, we will do it
# while we also transform the disjuncts.
or_expr = 0
for disjunct in obj.disjuncts:
or_expr += disjunct.indicator_var
self._transform_disjunct(disjunct, transBlock, varSet,
localVars[disjunct])
orConstraint.add(index, (or_expr, 1))
for i, var in enumerate(varSet):
disaggregatedExpr = 0
for disjunct in obj.disjuncts:
if 'chull' not in disjunct._gdp_transformation_info:
if not disjunct.indicator_var.is_fixed() \
or value(disjunct.indicator_var) != 0:
raise RuntimeError(
"GDP chull: disjunct was not relaxed, but "
"does not appear to be correctly deactivated.")
continue
disaggregatedVar = disjunct._gdp_transformation_info['chull'][
'disaggregatedVars'][var]
disaggregatedExpr += disaggregatedVar
if type(index) is tuple:
consIdx = index + (i, )
elif parent_component.is_indexed():
consIdx = (index, ) + (i, )
else:
consIdx = i
disaggregationConstraint.add(consIdx, var == disaggregatedExpr)
def _transform_disjunct(self, obj, transBlock, varSet, localVars):
if hasattr(obj, "_gdp_transformation_info"):
infodict = obj._gdp_transformation_info
# If the user has something with our name that is not a dict, we
# scream. If they have a dict with this name then we are just going
# to use it...
if type(infodict) is not dict:
raise GDP_Error(
"Disjunct %s contains an attribute named "
"_gdp_transformation_info. The transformation requires "
"that it can create this attribute!" % obj.name)
else:
infodict = obj._gdp_transformation_info = {}
# deactivated means either we've already transformed or user deactivated
if not obj.active:
if obj.indicator_var.is_fixed():
if value(obj.indicator_var) == 0:
# The user cleanly deactivated the disjunct: there
# is nothing for us to do here.
return
else:
raise GDP_Error(
"The disjunct %s is deactivated, but the "
"indicator_var is fixed to %s. This makes no sense." %
(obj.name, value(obj.indicator_var)))
if not infodict.get('relaxed', False):
raise GDP_Error(
"The disjunct %s is deactivated, but the "
"indicator_var is not fixed and the disjunct does not "
"appear to have been relaxed. This makes no sense." %
(obj.name, ))
if 'chull' in infodict:
# we've transformed it (with CHull), so don't do it again.
return
# add reference to original disjunct to info dict on
# transformation block
relaxedDisjuncts = transBlock.relaxedDisjuncts
relaxationBlock = relaxedDisjuncts[len(relaxedDisjuncts)]
relaxationBlockInfo = relaxationBlock._gdp_transformation_info = {
'src': obj,
'srcVars': ComponentMap(),
'srcConstraints': ComponentMap(),
'boundConstraintToSrcVar': ComponentMap(),
}
infodict['chull'] = chull = {
'relaxationBlock': relaxationBlock,
'relaxedConstraints': ComponentMap(),
'disaggregatedVars': ComponentMap(),
'bigmConstraints': ComponentMap(),
}
# if this is a disjunctData from an indexed disjunct, we are
# going to want to check at the end that the container is
# deactivated if everything in it is. So we save it in our
# dictionary of things to check if it isn't there already.
disjParent = obj.parent_component()
if disjParent.is_indexed() and \
disjParent not in transBlock.disjContainers:
transBlock.disjContainers.add(disjParent)
# add the disaggregated variables and their bigm constraints
# to the relaxationBlock
for var in varSet:
lb = var.lb
ub = var.ub
if lb is None or ub is None:
raise GDP_Error("Variables that appear in disjuncts must be "
"bounded in order to use the chull "
"transformation! Missing bound for %s." %
(var.name))
disaggregatedVar = Var(within=Reals,
bounds=(min(0, lb), max(0, ub)),
initialize=var.value)
# naming conflicts are possible here since this is a bunch
# of variables from different blocks coming together, so we
# get a unique name
disaggregatedVarName = unique_component_name(
relaxationBlock, var.local_name)
relaxationBlock.add_component(disaggregatedVarName,
disaggregatedVar)
chull['disaggregatedVars'][var] = disaggregatedVar
relaxationBlockInfo['srcVars'][disaggregatedVar] = var
bigmConstraint = Constraint(transBlock.lbub)
relaxationBlock.add_component(disaggregatedVarName + "_bounds",
bigmConstraint)
if lb:
bigmConstraint.add('lb',
obj.indicator_var * lb <= disaggregatedVar)
if ub:
bigmConstraint.add('ub',
disaggregatedVar <= obj.indicator_var * ub)
chull['bigmConstraints'][var] = bigmConstraint
relaxationBlockInfo['boundConstraintToSrcVar'][
bigmConstraint] = var
for var in localVars:
lb = var.lb
ub = var.ub
if lb is None or ub is None:
raise GDP_Error("Variables that appear in disjuncts must be "
"bounded in order to use the chull "
"transformation! Missing bound for %s." %
(var.name))
if value(lb) > 0:
var.setlb(0)
if value(ub) < 0:
var.setub(0)
# naming conflicts are possible here since this is a bunch
# of variables from different blocks coming together, so we
# get a unique name
conName = unique_component_name(relaxationBlock,
var.local_name + "_bounds")
bigmConstraint = Constraint(transBlock.lbub)
relaxationBlock.add_component(conName, bigmConstraint)
bigmConstraint.add('lb', obj.indicator_var * lb <= var)
bigmConstraint.add('ub', var <= obj.indicator_var * ub)
chull['bigmConstraints'][var] = bigmConstraint
relaxationBlockInfo['boundConstraintToSrcVar'][
bigmConstraint] = var
var_substitute_map = dict(
(id(v), newV) for v, newV in iteritems(chull['disaggregatedVars']))
zero_substitute_map = dict(
(id(v), NumericConstant(0))
for v, newV in iteritems(chull['disaggregatedVars']))
zero_substitute_map.update(
(id(v), NumericConstant(0)) for v in localVars)
# Transform each component within this disjunct
self._transform_block_components(obj, obj, infodict,
var_substitute_map,
zero_substitute_map)
# deactivate disjunct so we know we've relaxed it
obj._deactivate_without_fixing_indicator()
infodict['relaxed'] = True
def _transform_block_components(self, block, disjunct, infodict,
var_substitute_map, zero_substitute_map):
# Look through the component map of block and transform
# everything we have a handler for. Yell if we don't know how
# to handle it.
for name, obj in list(iteritems(block.component_map())):
if hasattr(obj, 'active') and not obj.active:
continue
handler = self.handlers.get(obj.type(), None)
if not handler:
if handler is None:
raise GDP_Error(
"No chull transformation handler registered "
"for modeling components of type %s" % obj.type())
continue
# obj is what we are transforming, we pass disjunct
# through so that we will have access to the indicator
# variables down the line.
handler(obj, disjunct, infodict, var_substitute_map,
zero_substitute_map)
def _warn_for_active_disjunction(self, disjunction, disjunct, infodict,
var_substitute_map, zero_substitute_map):
# this should only have gotten called if the disjunction is active
assert disjunction.active
problemdisj = disjunction
if disjunction.is_indexed():
for i in sorted(iterkeys(disjunction)):
if disjunction[i].active:
# a _DisjunctionData is active, we will yell about
# it specifically.
problemdisj = disjunction[i]
break
# None of the _DisjunctionDatas were actually active. We
# are OK and we can deactivate the container.
else:
disjunction.deactivate()
return
parentblock = problemdisj.parent_block()
# the disjunction should only have been active if it wasn't transformed
assert (not hasattr(parentblock, "_gdp_transformation_info")) or \
problemdisj.name not in parentblock._gdp_transformation_info
raise GDP_Error("Found untransformed disjunction %s in disjunct %s! "
"The disjunction must be transformed before the "
"disjunct. If you are using targets, put the "
"disjunction before the disjunct in the list." \
% (problemdisj.name, disjunct.name))
def _warn_for_active_disjunct(self, innerdisjunct, outerdisjunct, infodict,
var_substitute_map, zero_substitute_map):
assert innerdisjunct.active
problemdisj = innerdisjunct
if innerdisjunct.is_indexed():
for i in sorted(iterkeys(innerdisjunct)):
if innerdisjunct[i].active:
# This is shouldn't be true, we will complain about it.
problemdisj = innerdisjunct[i]
break
# None of the _DisjunctDatas were actually active, so we
# are fine and we can deactivate the container.
else:
# HACK: See above about _deactivate_without_fixing_indicator
ActiveComponent.deactivate(innerdisjunct)
return
raise GDP_Error(
"Found active disjunct {0} in disjunct {1}! Either {0} "
"is not in a disjunction or the disjunction it is in "
"has not been transformed. {0} needs to be deactivated "
"or its disjunction transformed before {1} can be "
"transformed.".format(problemdisj.name, outerdisjunct.name))
def _transform_block_on_disjunct(self, block, disjunct, infodict,
var_substitute_map, zero_substitute_map):
# We look through everything on the component map of the block
# and transform it just as we would if it was on the disjunct
# directly. (We are passing the disjunct through so that when
# we find constraints, _xform_constraint will have access to
# the correct indicator variable.
self._transform_block_components(block, disjunct, infodict,
var_substitute_map,
zero_substitute_map)
def _xform_constraint(self, obj, disjunct, infodict, var_substitute_map,
zero_substitute_map):
# we will put a new transformed constraint on the relaxation block.
relaxationBlock = infodict['chull']['relaxationBlock']
transBlock = relaxationBlock.parent_block()
varMap = infodict['chull']['disaggregatedVars']
# Though rare, it is possible to get naming conflicts here
# since constraints from all blocks are getting moved onto the
# same block. So we get a unique name
name = unique_component_name(relaxationBlock, obj.name)
if obj.is_indexed():
try:
newConstraint = Constraint(obj.index_set(), transBlock.lbub)
except:
# The original constraint may have been indexed by a
# non-concrete set (like an Any). We will give up on
# strict index verification and just blindly proceed.
newConstraint = Constraint(Any)
else:
newConstraint = Constraint(transBlock.lbub)
relaxationBlock.add_component(name, newConstraint)
# add mapping of original constraint to transformed constraint
# in transformation info dictionary
infodict['chull']['relaxedConstraints'][obj] = newConstraint
# add mapping of transformed constraint back to original constraint (we
# know that the info dict is already created because this only got
# called if we were transforming a disjunct...)
relaxationBlock._gdp_transformation_info['srcConstraints'][
newConstraint] = obj
for i in sorted(iterkeys(obj)):
c = obj[i]
if not c.active:
continue
NL = c.body.polynomial_degree() not in (0, 1)
EPS = self._config.EPS
mode = self._config.perspective_function
# We need to evaluate the expression at the origin *before*
# we substitute the expression variables with the
# disaggregated variables
if not NL or mode == "FurmanSawayaGrossmann":
h_0 = clone_without_expression_components(
c.body, substitute=zero_substitute_map)
y = disjunct.indicator_var
if NL:
if mode == "LeeGrossmann":
sub_expr = clone_without_expression_components(
c.body,
substitute=dict(
(var, subs / y)
for var, subs in iteritems(var_substitute_map)))
expr = sub_expr * y
elif mode == "GrossmannLee":
sub_expr = clone_without_expression_components(
c.body,
substitute=dict(
(var, subs / (y + EPS))
for var, subs in iteritems(var_substitute_map)))
expr = (y + EPS) * sub_expr
elif mode == "FurmanSawayaGrossmann":
sub_expr = clone_without_expression_components(
c.body,
substitute=dict(
(var, subs / ((1 - EPS) * y + EPS))
for var, subs in iteritems(var_substitute_map)))
expr = (
(1 - EPS) * y + EPS) * sub_expr - EPS * h_0 * (1 - y)
else:
raise RuntimeError("Unknown NL CHull mode")
else:
expr = clone_without_expression_components(
c.body, substitute=var_substitute_map)
if c.equality:
if NL:
newConsExpr = expr == c.lower * y
else:
v = list(EXPR.identify_variables(expr))
if len(v) == 1 and not c.lower:
# Setting a variable to 0 in a disjunct is
# *very* common. We should recognize that in
# that structure, the disaggregated variable
# will also be fixed to 0.
v[0].fix(0)
continue
newConsExpr = expr - (1 - y) * h_0 == c.lower * y
if obj.is_indexed():
newConstraint.add((i, 'eq'), newConsExpr)
else:
newConstraint.add('eq', newConsExpr)
continue
if c.lower is not None:
# TODO: At the moment there is no reason for this to be in both
# lower and upper... I think there could be though if I say what
# the new constraint is going to be or something.
if __debug__ and logger.isEnabledFor(logging.DEBUG):
logger.debug(
"GDP(cHull): Transforming constraint " + "'%s'",
c.name)
if NL:
newConsExpr = expr >= c.lower * y
else:
newConsExpr = expr - (1 - y) * h_0 >= c.lower * y
if obj.is_indexed():
newConstraint.add((i, 'lb'), newConsExpr)
else:
newConstraint.add('lb', newConsExpr)
if c.upper is not None:
if __debug__ and logger.isEnabledFor(logging.DEBUG):
logger.debug(
"GDP(cHull): Transforming constraint " + "'%s'",
c.name)
if NL:
newConsExpr = expr <= c.upper * y
else:
newConsExpr = expr - (1 - y) * h_0 <= c.upper * y
if obj.is_indexed():
newConstraint.add((i, 'ub'), newConsExpr)
else:
newConstraint.add('ub', newConsExpr)
class Hull_Reformulation(Transformation):
"""Relax disjunctive model by forming the hull reformulation.
Relaxes a disjunctive model into an algebraic model by forming the
hull reformulation of each disjunction.
This transformation accepts the following keyword arguments:
Parameters
----------
perspective_function : str
The perspective function used for the disaggregated variables.
Must be one of 'FurmanSawayaGrossmann' (default),
'LeeGrossmann', or 'GrossmannLee'
EPS : float
The value to use for epsilon [default: 1e-4]
targets : (block, disjunction, or list of those types)
The targets to transform. This can be a block, disjunction, or a
list of blocks and Disjunctions [default: the instance]
The transformation will create a new Block with a unique
name beginning "_pyomo_gdp_hull_reformulation". That Block will
contain an indexed Block named "relaxedDisjuncts", which will hold
the relaxed disjuncts. This block is indexed by an integer
indicating the order in which the disjuncts were relaxed.
Each block has a dictionary "_constraintMap":
'srcConstraints': ComponentMap(<transformed constraint>:
<src constraint>),
'transformedConstraints':ComponentMap(<src constraint container>:
<transformed constraint container>,
<src constraintData>:
[<transformed constraintDatas>]
)
It will have a dictionary "_disaggregatedVarMap:
'srcVar': ComponentMap(<src var>:<disaggregated var>),
'disaggregatedVar': ComponentMap(<disaggregated var>:<src var>)
And, last, it will have a ComponentMap "_bigMConstraintMap":
<disaggregated var>:<bounds constraint>
All transformed Disjuncts will have a pointer to the block their transformed
constraints are on, and all transformed Disjunctions will have a
pointer to the corresponding OR or XOR constraint.
The _pyomo_gdp_hull_reformulation block will have a ComponentMap
"_disaggregationConstraintMap":
<src var>:ComponentMap(<srcDisjunction>: <disaggregation constraint>)
"""
CONFIG = cfg.ConfigBlock('gdp.hull')
CONFIG.declare(
'targets',
cfg.ConfigValue(
default=None,
domain=target_list,
description="target or list of targets that will be relaxed",
doc="""
This specifies the target or list of targets to relax as either a
component or a list of components. If None (default), the entire model
is transformed. Note that if the transformation is done out of place,
the list of targets should be attached to the model before it is cloned,
and the list will specify the targets on the cloned instance."""))
CONFIG.declare(
'perspective function',
cfg.ConfigValue(
default='FurmanSawayaGrossmann',
domain=cfg.In(
['FurmanSawayaGrossmann', 'LeeGrossmann', 'GrossmannLee']),
description='perspective function used for variable disaggregation',
doc="""
The perspective function used for variable disaggregation
"LeeGrossmann" is the original NL convex hull from Lee &
Grossmann (2000) [1]_, which substitutes nonlinear constraints
h_ik(x) <= 0
with
x_k = sum( nu_ik )
y_ik * h_ik( nu_ik/y_ik ) <= 0
"GrossmannLee" is an updated formulation from Grossmann &
Lee (2003) [2]_, which avoids divide-by-0 errors by using:
x_k = sum( nu_ik )
(y_ik + eps) * h_ik( nu_ik/(y_ik + eps) ) <= 0
"FurmanSawayaGrossmann" (default) is an improved relaxation [3]_
that is exact at 0 and 1 while avoiding numerical issues from
the Lee & Grossmann formulation by using:
x_k = sum( nu_ik )
((1-eps)*y_ik + eps) * h_ik( nu_ik/((1-eps)*y_ik + eps) ) \
- eps * h_ki(0) * ( 1-y_ik ) <= 0
References
----------
.. [1] Lee, S., & Grossmann, I. E. (2000). New algorithms for
nonlinear generalized disjunctive programming. Computers and
Chemical Engineering, 24, 2125-2141
.. [2] Grossmann, I. E., & Lee, S. (2003). Generalized disjunctive
programming: Nonlinear convex hull relaxation and algorithms.
Computational Optimization and Applications, 26, 83-100.
.. [3] Furman, K., Sawaya, N., and Grossmann, I. A computationally
useful algebraic representation of nonlinear disjunctive convex
sets using the perspective function. Optimization Online
(2016). http://www.optimization-online.org/DB_HTML/2016/07/5544.html.
"""))
CONFIG.declare(
'EPS',
cfg.ConfigValue(
default=1e-4,
domain=cfg.PositiveFloat,
description="Epsilon value to use in perspective function",
))
CONFIG.declare(
'assume_fixed_vars_permanent',
cfg.ConfigValue(
default=False,
domain=bool,
description=
"Boolean indicating whether or not to transform so that the "
"the transformed model will still be valid when fixed Vars are unfixed.",
doc="""
If True, the transformation will not disaggregate fixed variables.
This means that if a fixed variable is unfixed after transformation,
the transformed model is no longer valid. By default, the transformation
will disagregate fixed variables so that any later fixing and unfixing
will be valid in the transformed model.
"""))
def __init__(self):
super(Hull_Reformulation, self).__init__()
self.handlers = {
Constraint: self._transform_constraint,
Var: False,
Connector: False,
Expression: False,
Param: False,
Set: False,
SetOf: False,
RangeSet: False,
Suffix: False,
Disjunction: self._warn_for_active_disjunction,
Disjunct: self._warn_for_active_disjunct,
Block: self._transform_block_on_disjunct,
}
def _add_local_vars(self, block, local_var_dict):
localVars = block.component('LocalVars')
if type(localVars) is Suffix:
for disj, var_list in iteritems(localVars):
if local_var_dict.get(disj) is None:
local_var_dict[disj] = ComponentSet(var_list)
else:
local_var_dict[disj].update(var_list)
def _get_local_var_suffixes(self, block, local_var_dict):
# You can specify suffixes on any block (disjuncts included). This method
# starts from a Disjunct (presumably) and checks for a LocalVar suffixes
# going both up and down the tree, adding them into the dictionary that
# is the second argument.
# first look beneath where we are (there could be Blocks on this
# disjunct)
for b in block.component_data_objects(
Block,
descend_into=(Block),
active=True,
sort=SortComponents.deterministic):
self._add_local_vars(b, local_var_dict)
# now traverse upwards and get what's above
while block is not None:
self._add_local_vars(block, local_var_dict)
block = block.parent_block()
return local_var_dict
def _apply_to(self, instance, **kwds):
assert not NAME_BUFFER
try:
self._apply_to_impl(instance, **kwds)
finally:
# Clear the global name buffer now that we are done
NAME_BUFFER.clear()
def _apply_to_impl(self, instance, **kwds):
self._config = self.CONFIG(kwds.pop('options', {}))
self._config.set_value(kwds)
targets = self._config.targets
if targets is None:
targets = (instance, )
knownBlocks = {}
for t in targets:
# check that t is in fact a child of instance
if not is_child_of(
parent=instance, child=t, knownBlocks=knownBlocks):
raise GDP_Error(
"Target '%s' is not a component on instance '%s'!" %
(t.name, instance.name))
elif t.ctype is Disjunction:
if t.is_indexed():
self._transform_disjunction(t)
else:
self._transform_disjunctionData(t, t.index())
elif t.ctype in (Block, Disjunct):
if t.is_indexed():
self._transform_block(t)
else:
self._transform_blockData(t)
else:
raise GDP_Error(
"Target '%s' was not a Block, Disjunct, or Disjunction. "
"It was of type %s and can't be transformed." %
(t.name, type(t)))
def _add_transformation_block(self, instance):
# make a transformation block on instance where we will store
# transformed components
transBlockName = unique_component_name(
instance, '_pyomo_gdp_hull_reformulation')
transBlock = Block()
instance.add_component(transBlockName, transBlock)
transBlock.relaxedDisjuncts = Block(NonNegativeIntegers)
transBlock.lbub = Set(initialize=['lb', 'ub', 'eq'])
# We will store all of the disaggregation constraints for any
# Disjunctions we transform onto this block here.
transBlock.disaggregationConstraints = Constraint(
NonNegativeIntegers, Any)
# This will map from srcVar to a map of srcDisjunction to the
# disaggregation constraint corresponding to srcDisjunction
transBlock._disaggregationConstraintMap = ComponentMap()
return transBlock
def _transform_block(self, obj):
for i in sorted(iterkeys(obj)):
self._transform_blockData(obj[i])
def _transform_blockData(self, obj):
# Transform every (active) disjunction in the block
for disjunction in obj.component_objects(
Disjunction,
active=True,
sort=SortComponents.deterministic,
descend_into=(Block, Disjunct),
descent_order=TraversalStrategy.PostfixDFS):
self._transform_disjunction(disjunction)
def _add_xor_constraint(self, disjunction, transBlock):
# Put XOR constraint on the transformation block
# We never do this for just a DisjunctionData because we need
# to know about the index set of its parent component. So if
# we called this on a DisjunctionData, we did something wrong.
assert isinstance(disjunction, Disjunction)
# check if the constraint already exists
if disjunction._algebraic_constraint is not None:
return disjunction._algebraic_constraint()
# add the XOR (or OR) constraints to parent block (with
# unique name) It's indexed if this is an
# IndexedDisjunction, not otherwise
orC = Constraint(disjunction.index_set())
transBlock.add_component(
unique_component_name(transBlock,
disjunction.getname(fully_qualified=True,
name_buffer=NAME_BUFFER) +\
'_xor'), orC)
disjunction._algebraic_constraint = weakref_ref(orC)
return orC
def _transform_disjunction(self, obj):
# NOTE: this check is actually necessary because it's possible we go
# straight to this function when we use targets.
if not obj.active:
return
# put the transformation block on the parent block of the Disjunction,
# unless this is a disjunction we have seen in a prior call to hull, in
# which case we will use the same transformation block we created
# before.
if obj._algebraic_constraint is not None:
transBlock = obj._algebraic_constraint().parent_block()
else:
transBlock = self._add_transformation_block(obj.parent_block())
# and create the xor constraint
xorConstraint = self._add_xor_constraint(obj, transBlock)
# create the disjunction constraint and disaggregation
# constraints and then relax each of the disjunctionDatas
for i in sorted(iterkeys(obj)):
self._transform_disjunctionData(obj[i], i, transBlock)
# deactivate so the writers will be happy
obj.deactivate()
def _transform_disjunctionData(self, obj, index, transBlock=None):
if not obj.active:
return
# Hull reformulation doesn't work if this is an OR constraint. So if
# xor is false, give up
if not obj.xor:
raise GDP_Error("Cannot do hull reformulation for "
"Disjunction '%s' with OR constraint. "
"Must be an XOR!" % obj.name)
if transBlock is None:
# It's possible that we have already created a transformation block
# for another disjunctionData from this same container. If that's
# the case, let's use the same transformation block. (Else it will
# be really confusing that the XOR constraint goes to that old block
# but we create a new one here.)
if obj.parent_component()._algebraic_constraint is not None:
transBlock = obj.parent_component()._algebraic_constraint().\
parent_block()
else:
transBlock = self._add_transformation_block(obj.parent_block())
parent_component = obj.parent_component()
orConstraint = self._add_xor_constraint(parent_component, transBlock)
disaggregationConstraint = transBlock.disaggregationConstraints
disaggregationConstraintMap = transBlock._disaggregationConstraintMap
# Just because it's unlikely this is what someone meant to do...
if len(obj.disjuncts) == 0:
raise GDP_Error(
"Disjunction '%s' is empty. This is "
"likely indicative of a modeling error." %
obj.getname(fully_qualified=True, name_buffer=NAME_BUFFER))
# We first go through and collect all the variables that we
# are going to disaggregate.
varOrder_set = ComponentSet()
varOrder = []
varsByDisjunct = ComponentMap()
localVarsByDisjunct = ComponentMap()
include_fixed_vars = not self._config.assume_fixed_vars_permanent
for disjunct in obj.disjuncts:
disjunctVars = varsByDisjunct[disjunct] = ComponentSet()
for cons in disjunct.component_data_objects(
Constraint,
active=True,
sort=SortComponents.deterministic,
descend_into=Block):
# [ESJ 02/14/2020] By default, we disaggregate fixed variables
# on the philosophy that fixing is not a promise for the future
# and we are mathematically wrong if we don't transform these
# correctly and someone later unfixes them and keeps playing
# with their transformed model. However, the user may have set
# assume_fixed_vars_permanent to True in which case we will skip
# them
for var in EXPR.identify_variables(
cons.body, include_fixed=include_fixed_vars):
# Note the use of a list so that we will
# eventually disaggregate the vars in a
# deterministic order (the order that we found
# them)
disjunctVars.add(var)
if not var in varOrder_set:
varOrder.append(var)
varOrder_set.add(var)
# check for LocalVars Suffix
localVarsByDisjunct = self._get_local_var_suffixes(
disjunct, localVarsByDisjunct)
# We will disaggregate all variables which are not explicitly declared
# as being local. Note however, that we do declare our own disaggregated
# variables as local, so they will not be re-disaggregated.
varSet = []
# values of localVarsByDisjunct are ComponentSets, so we need this for
# determinism (we iterate through the localVars later)
localVars = []
for var in varOrder:
disjuncts = [d for d in varsByDisjunct if var in varsByDisjunct[d]]
# clearly not local if used in more than one disjunct
if len(disjuncts) > 1:
if __debug__ and logger.isEnabledFor(logging.DEBUG):
logger.debug("Assuming '%s' is not a local var since it is"
"used in multiple disjuncts." %
var.getname(fully_qualified=True,
name_buffer=NAME_BUFFER))
varSet.append(var)
elif localVarsByDisjunct.get(disjuncts[0]) is not None:
if var in localVarsByDisjunct[disjuncts[0]]:
localVars.append(var)
else:
varSet.append(var)
else:
varSet.append(var)
# Now that we know who we need to disaggregate, we will do it
# while we also transform the disjuncts.
or_expr = 0
for disjunct in obj.disjuncts:
or_expr += disjunct.indicator_var
self._transform_disjunct(disjunct, transBlock, varSet, localVars)
orConstraint.add(index, (or_expr, 1))
# map the DisjunctionData to its XOR constraint to mark it as
# transformed
obj._algebraic_constraint = weakref_ref(orConstraint[index])
for i, var in enumerate(varSet):
disaggregatedExpr = 0
for disjunct in obj.disjuncts:
if disjunct._transformation_block is None:
# Because we called _transform_disjunct in the loop above,
# we know that if this isn't transformed it is because it
# was cleanly deactivated, and we can just skip it.
continue
disaggregatedVar = disjunct._transformation_block().\
_disaggregatedVarMap['disaggregatedVar'][var]
disaggregatedExpr += disaggregatedVar
disaggregationConstraint.add((i, index), var == disaggregatedExpr)
# and update the map so that we can find this later. We index by
# variable and the particular disjunction because there is a
# different one for each disjunction
if disaggregationConstraintMap.get(var) is not None:
disaggregationConstraintMap[var][
obj] = disaggregationConstraint[(i, index)]
else:
thismap = disaggregationConstraintMap[var] = ComponentMap()
thismap[obj] = disaggregationConstraint[(i, index)]
# deactivate for the writers
obj.deactivate()
def _transform_disjunct(self, obj, transBlock, varSet, localVars):
# deactivated should only come from the user
if not obj.active:
if obj.indicator_var.is_fixed():
if value(obj.indicator_var) == 0:
# The user cleanly deactivated the disjunct: there
# is nothing for us to do here.
return
else:
raise GDP_Error(
"The disjunct '%s' is deactivated, but the "
"indicator_var is fixed to %s. This makes no sense." %
(obj.name, value(obj.indicator_var)))
if obj._transformation_block is None:
raise GDP_Error(
"The disjunct '%s' is deactivated, but the "
"indicator_var is not fixed and the disjunct does not "
"appear to have been relaxed. This makes no sense. "
"(If the intent is to deactivate the disjunct, fix its "
"indicator_var to 0.)" % (obj.name, ))
if obj._transformation_block is not None:
# we've transformed it, which means this is the second time it's
# appearing in a Disjunction
raise GDP_Error(
"The disjunct '%s' has been transformed, but a disjunction "
"it appears in has not. Putting the same disjunct in "
"multiple disjunctions is not supported." % obj.name)
# create a relaxation block for this disjunct
relaxedDisjuncts = transBlock.relaxedDisjuncts
relaxationBlock = relaxedDisjuncts[len(relaxedDisjuncts)]
relaxationBlock.localVarReferences = Block()
# Put the disaggregated variables all on their own block so that we can
# isolate the name collisions and still have complete control over the
# names on this block. (This is for peace of mind now, but will matter
# in the future for adding the binaries corresponding to Boolean
# indicator vars.)
relaxationBlock.disaggregatedVars = Block()
# add the map that will link back and forth between transformed
# constraints and their originals.
relaxationBlock._constraintMap = {
'srcConstraints': ComponentMap(),
'transformedConstraints': ComponentMap()
}
# Map between disaggregated variables for this disjunct and their
# originals
relaxationBlock._disaggregatedVarMap = {
'srcVar': ComponentMap(),
'disaggregatedVar': ComponentMap(),
}
# Map between disaggregated variables and their lb*indicator <= var <=
# ub*indicator constraints
relaxationBlock._bigMConstraintMap = ComponentMap()
# add mappings to source disjunct (so we'll know we've relaxed)
obj._transformation_block = weakref_ref(relaxationBlock)
relaxationBlock._srcDisjunct = weakref_ref(obj)
# add Suffix to the relaxation block that disaggregated variables are
# local (in case this is nested in another Disjunct)
local_var_set = None
parent_disjunct = obj.parent_block()
while parent_disjunct is not None:
if parent_disjunct.ctype is Disjunct:
break
parent_disjunct = parent_disjunct.parent_block()
if parent_disjunct is not None:
localVarSuffix = relaxationBlock.LocalVars = Suffix(
direction=Suffix.LOCAL)
local_var_set = localVarSuffix[parent_disjunct] = ComponentSet()
# add the disaggregated variables and their bigm constraints
# to the relaxationBlock
for var in varSet:
lb = var.lb
ub = var.ub
if lb is None or ub is None:
raise GDP_Error("Variables that appear in disjuncts must be "
"bounded in order to use the hull "
"transformation! Missing bound for %s." %
(var.name))
disaggregatedVar = Var(within=Reals,
bounds=(min(0, lb), max(0, ub)),
initialize=var.value)
# naming conflicts are possible here since this is a bunch
# of variables from different blocks coming together, so we
# get a unique name
disaggregatedVarName = unique_component_name(
relaxationBlock.disaggregatedVars,
var.getname(fully_qualified=False, name_buffer=NAME_BUFFER),
)
relaxationBlock.disaggregatedVars.add_component(
disaggregatedVarName, disaggregatedVar)
# mark this as local because we won't re-disaggregate if this is a
# nested disjunction
if local_var_set is not None:
local_var_set.add(disaggregatedVar)
# store the mappings from variables to their disaggregated selves on
# the transformation block.
relaxationBlock._disaggregatedVarMap['disaggregatedVar'][
var] = disaggregatedVar
relaxationBlock._disaggregatedVarMap['srcVar'][
disaggregatedVar] = var
bigmConstraint = Constraint(transBlock.lbub)
relaxationBlock.add_component(disaggregatedVarName + "_bounds",
bigmConstraint)
if lb:
bigmConstraint.add('lb',
obj.indicator_var * lb <= disaggregatedVar)
if ub:
bigmConstraint.add('ub',
disaggregatedVar <= obj.indicator_var * ub)
relaxationBlock._bigMConstraintMap[
disaggregatedVar] = bigmConstraint
for var in localVars:
lb = var.lb
ub = var.ub
if lb is None or ub is None:
raise GDP_Error("Variables that appear in disjuncts must be "
"bounded in order to use the hull "
"transformation! Missing bound for %s." %
(var.name))
if value(lb) > 0:
var.setlb(0)
if value(ub) < 0:
var.setub(0)
# map it to itself
relaxationBlock._disaggregatedVarMap['disaggregatedVar'][var] = var
relaxationBlock._disaggregatedVarMap['srcVar'][var] = var
# naming conflicts are possible here since this is a bunch
# of variables from different blocks coming together, so we
# get a unique name
conName = unique_component_name(
relaxationBlock,
var.getname(fully_qualified=False, name_buffer=NAME_BUFFER) + \
"_bounds")
bigmConstraint = Constraint(transBlock.lbub)
relaxationBlock.add_component(conName, bigmConstraint)
if lb:
bigmConstraint.add('lb', obj.indicator_var * lb <= var)
if ub:
bigmConstraint.add('ub', var <= obj.indicator_var * ub)
relaxationBlock._bigMConstraintMap[var] = bigmConstraint
var_substitute_map = dict((id(v), newV) for v, newV in iteritems(
relaxationBlock._disaggregatedVarMap['disaggregatedVar']))
zero_substitute_map = dict((id(v), ZeroConstant) for v, newV in \
iteritems(
relaxationBlock._disaggregatedVarMap[
'disaggregatedVar']))
zero_substitute_map.update((id(v), ZeroConstant) for v in localVars)
# Transform each component within this disjunct
self._transform_block_components(obj, obj, var_substitute_map,
zero_substitute_map)
# deactivate disjunct so writers can be happy
obj._deactivate_without_fixing_indicator()
def _transform_block_components(self, block, disjunct, var_substitute_map,
zero_substitute_map):
# As opposed to bigm, in hull the only special thing we need to do for
# nested Disjunctions is to make sure that we move up local var
# references and also references to the disaggregated variables so that
# all will be accessible after we transform this
# Disjunct. Mathematically, there is nothing to do: the indicator
# variables and disaggregated variables of the inner disjunction will
# need to be disaggregated again anyway, and nothing will get
# double-bigm-ed. (If an untransformed disjunction is lurking here, we
# will catch it below).
# add references to all local variables on block (including the
# indicator_var)
disjunctBlock = disjunct._transformation_block()
varRefBlock = disjunctBlock.localVarReferences
for v in block.component_objects(Var, descend_into=Block, active=None):
varRefBlock.add_component(
unique_component_name(
varRefBlock,
v.getname(fully_qualified=True, name_buffer=NAME_BUFFER)),
Reference(v))
destinationBlock = disjunctBlock.parent_block()
for obj in block.component_data_objects(
Disjunction,
sort=SortComponents.deterministic,
descend_into=(Block)):
if obj.algebraic_constraint is None:
# This could be bad if it's active since that means its
# untransformed, but we'll wait to yell until the next loop
continue
# get this disjunction's relaxation block.
transBlock = obj.algebraic_constraint().parent_block()
self._transfer_var_references(transBlock, destinationBlock)
# Look through the component map of block and transform everything we
# have a handler for. Yell if we don't know how to handle it. (Note that
# because we only iterate through active components, this means
# non-ActiveComponent types cannot have handlers.)
for obj in block.component_objects(active=True, descend_into=False):
handler = self.handlers.get(obj.ctype, None)
if not handler:
if handler is None:
raise GDP_Error(
"No hull transformation handler registered "
"for modeling components of type %s. If your "
"disjuncts contain non-GDP Pyomo components that "
"require transformation, please transform them first."
% obj.ctype)
continue
# obj is what we are transforming, we pass disjunct
# through so that we will have access to the indicator
# variables down the line.
handler(obj, disjunct, var_substitute_map, zero_substitute_map)
def _transfer_var_references(self, fromBlock, toBlock):
disjunctList = toBlock.relaxedDisjuncts
for idx, disjunctBlock in iteritems(fromBlock.relaxedDisjuncts):
# move all the of the local var references
newblock = disjunctList[len(disjunctList)]
newblock.localVarReferences = Block()
newblock.localVarReferences.transfer_attributes_from(
disjunctBlock.localVarReferences)
def _warn_for_active_disjunction(self, disjunction, disjunct,
var_substitute_map, zero_substitute_map):
_warn_for_active_disjunction(disjunction, disjunct, NAME_BUFFER)
def _warn_for_active_disjunct(self, innerdisjunct, outerdisjunct,
var_substitute_map, zero_substitute_map):
_warn_for_active_disjunct(innerdisjunct, outerdisjunct, NAME_BUFFER)
def _transform_block_on_disjunct(self, block, disjunct, var_substitute_map,
zero_substitute_map):
# We look through everything on the component map of the block
# and transform it just as we would if it was on the disjunct
# directly. (We are passing the disjunct through so that when
# we find constraints, _transform_constraint will have access to
# the correct indicator variable.
for i in sorted(iterkeys(block)):
self._transform_block_components(block[i], disjunct,
var_substitute_map,
zero_substitute_map)
def _transform_constraint(self, obj, disjunct, var_substitute_map,
zero_substitute_map):
# we will put a new transformed constraint on the relaxation block.
relaxationBlock = disjunct._transformation_block()
transBlock = relaxationBlock.parent_block()
varMap = relaxationBlock._disaggregatedVarMap['disaggregatedVar']
constraintMap = relaxationBlock._constraintMap
# Though rare, it is possible to get naming conflicts here
# since constraints from all blocks are getting moved onto the
# same block. So we get a unique name
name = unique_component_name(
relaxationBlock,
obj.getname(fully_qualified=True, name_buffer=NAME_BUFFER))
if obj.is_indexed():
newConstraint = Constraint(obj.index_set(), transBlock.lbub)
else:
newConstraint = Constraint(transBlock.lbub)
relaxationBlock.add_component(name, newConstraint)
# map the containers:
# add mapping of original constraint to transformed constraint
if obj.is_indexed():
constraintMap['transformedConstraints'][obj] = newConstraint
# add mapping of transformed constraint container back to original
# constraint container (or SimpleConstraint)
constraintMap['srcConstraints'][newConstraint] = obj
for i in sorted(iterkeys(obj)):
c = obj[i]
if not c.active:
continue
NL = c.body.polynomial_degree() not in (0, 1)
EPS = self._config.EPS
mode = self._config.perspective_function
# We need to evaluate the expression at the origin *before*
# we substitute the expression variables with the
# disaggregated variables
if not NL or mode == "FurmanSawayaGrossmann":
h_0 = clone_without_expression_components(
c.body, substitute=zero_substitute_map)
y = disjunct.indicator_var
if NL:
if mode == "LeeGrossmann":
sub_expr = clone_without_expression_components(
c.body,
substitute=dict(
(var, subs / y)
for var, subs in iteritems(var_substitute_map)))
expr = sub_expr * y
elif mode == "GrossmannLee":
sub_expr = clone_without_expression_components(
c.body,
substitute=dict(
(var, subs / (y + EPS))
for var, subs in iteritems(var_substitute_map)))
expr = (y + EPS) * sub_expr
elif mode == "FurmanSawayaGrossmann":
sub_expr = clone_without_expression_components(
c.body,
substitute=dict(
(var, subs / ((1 - EPS) * y + EPS))
for var, subs in iteritems(var_substitute_map)))
expr = (
(1 - EPS) * y + EPS) * sub_expr - EPS * h_0 * (1 - y)
else:
raise RuntimeError("Unknown NL Hull mode")
else:
expr = clone_without_expression_components(
c.body, substitute=var_substitute_map)
if c.equality:
if NL:
# ESJ TODO: This can't happen right? This is the only
# obvious case where someone has messed up, but this has to
# be nonconvex, right? Shouldn't we tell them?
newConsExpr = expr == c.lower * y
else:
v = list(EXPR.identify_variables(expr))
if len(v) == 1 and not c.lower:
# Setting a variable to 0 in a disjunct is
# *very* common. We should recognize that in
# that structure, the disaggregated variable
# will also be fixed to 0.
v[0].fix(0)
# ESJ: If you ask where the transformed constraint is,
# the answer is nowhere. Really, it is in the bounds of
# this variable, so I'm going to return
# it. Alternatively we could return an empty list, but I
# think I like this better.
constraintMap['transformedConstraints'][c] = [v[0]]
# Reverse map also (this is strange)
constraintMap['srcConstraints'][v[0]] = c
continue
newConsExpr = expr - (1 - y) * h_0 == c.lower * y
if obj.is_indexed():
newConstraint.add((i, 'eq'), newConsExpr)
# map the _ConstraintDatas (we mapped the container above)
constraintMap['transformedConstraints'][c] = [
newConstraint[i, 'eq']
]
constraintMap['srcConstraints'][newConstraint[i, 'eq']] = c
else:
newConstraint.add('eq', newConsExpr)
# map to the _ConstraintData (And yes, for
# SimpleConstraints, this is overwriting the map to the
# container we made above, and that is what I want to
# happen. SimpleConstraints will map to lists. For
# IndexedConstraints, we can map the container to the
# container, but more importantly, we are mapping the
# _ConstraintDatas to each other above)
constraintMap['transformedConstraints'][c] = [
newConstraint['eq']
]
constraintMap['srcConstraints'][newConstraint['eq']] = c
continue
if c.lower is not None:
if __debug__ and logger.isEnabledFor(logging.DEBUG):
_name = c.getname(fully_qualified=True,
name_buffer=NAME_BUFFER)
logger.debug(
"GDP(Hull): Transforming constraint " + "'%s'", _name)
if NL:
newConsExpr = expr >= c.lower * y
else:
newConsExpr = expr - (1 - y) * h_0 >= c.lower * y
if obj.is_indexed():
newConstraint.add((i, 'lb'), newConsExpr)
constraintMap['transformedConstraints'][c] = [
newConstraint[i, 'lb']
]
constraintMap['srcConstraints'][newConstraint[i, 'lb']] = c
else:
newConstraint.add('lb', newConsExpr)
constraintMap['transformedConstraints'][c] = [
newConstraint['lb']
]
constraintMap['srcConstraints'][newConstraint['lb']] = c
if c.upper is not None:
if __debug__ and logger.isEnabledFor(logging.DEBUG):
_name = c.getname(fully_qualified=True,
name_buffer=NAME_BUFFER)
logger.debug(
"GDP(Hull): Transforming constraint " + "'%s'", _name)
if NL:
newConsExpr = expr <= c.upper * y
else:
newConsExpr = expr - (1 - y) * h_0 <= c.upper * y
if obj.is_indexed():
newConstraint.add((i, 'ub'), newConsExpr)
# map (have to account for fact we might have created list
# above
transformed = constraintMap['transformedConstraints'].get(
c)
if transformed is not None:
transformed.append(newConstraint[i, 'ub'])
else:
constraintMap['transformedConstraints'][c] = [
newConstraint[i, 'ub']
]
constraintMap['srcConstraints'][newConstraint[i, 'ub']] = c
else:
newConstraint.add('ub', newConsExpr)
transformed = constraintMap['transformedConstraints'].get(
c)
if transformed is not None:
transformed.append(newConstraint['ub'])
else:
constraintMap['transformedConstraints'][c] = [
newConstraint['ub']
]
constraintMap['srcConstraints'][newConstraint['ub']] = c
# deactivate now that we have transformed
obj.deactivate()
# These are all functions to retrieve transformed components from
# original ones and vice versa.
@wraps(get_src_disjunct)
def get_src_disjunct(self, transBlock):
return get_src_disjunct(transBlock)
@wraps(get_src_disjunction)
def get_src_disjunction(self, xor_constraint):
return get_src_disjunction(xor_constraint)
@wraps(get_src_constraint)
def get_src_constraint(self, transformedConstraint):
return get_src_constraint(transformedConstraint)
@wraps(get_transformed_constraints)
def get_transformed_constraints(self, srcConstraint):
return get_transformed_constraints(srcConstraint)
def get_disaggregated_var(self, v, disjunct):
"""
Returns the disaggregated variable corresponding to the Var v and the
Disjunct disjunct.
If v is a local variable, this method will return v.
Parameters
----------
v: a Var which appears in a constraint in a transformed Disjunct
disjunct: a transformed Disjunct in which v appears
"""
if disjunct._transformation_block is None:
raise GDP_Error("Disjunct '%s' has not been transformed" %
disjunct.name)
transBlock = disjunct._transformation_block()
try:
return transBlock._disaggregatedVarMap['disaggregatedVar'][v]
except:
logger.error("It does not appear '%s' is a "
"variable which appears in disjunct '%s'" %
(v.name, disjunct.name))
raise
def get_src_var(self, disaggregated_var):
"""
Returns the original model variable to which disaggregated_var
corresponds.
Parameters
----------
disaggregated_var: a Var which was created by the hull
transformation as a disaggregated variable
(and so appears on a transformation block
of some Disjunct)
"""
transBlock = disaggregated_var.parent_block()
try:
return transBlock.parent_block(
)._disaggregatedVarMap['srcVar'][disaggregated_var]
except:
logger.error(
"'%s' does not appear to be a disaggregated variable" %
disaggregated_var.name)
raise
# retrieves the disaggregation constraint for original_var resulting from
# transforming disjunction
def get_disaggregation_constraint(self, original_var, disjunction):
"""
Returns the disaggregation (re-aggregation?) constraint
(which links the disaggregated variables to their original)
corresponding to original_var and the transformation of disjunction.
Parameters
----------
original_var: a Var which was disaggregated in the transformation
of Disjunction disjunction
disjunction: a transformed Disjunction containing original_var
"""
for disjunct in disjunction.disjuncts:
transBlock = disjunct._transformation_block
if transBlock is not None:
break
if transBlock is None:
raise GDP_Error(
"Disjunction '%s' has not been properly transformed: "
"None of its disjuncts are transformed." % disjunction.name)
try:
return transBlock().parent_block(
)._disaggregationConstraintMap[original_var][disjunction]
except:
logger.error("It doesn't appear that '%s' is a variable that was "
"disaggregated by Disjunction '%s'" %
(original_var.name, disjunction.name))
raise
def get_var_bounds_constraint(self, v):
"""
Returns the IndexedConstraint which sets a disaggregated
variable to be within its bounds when its Disjunct is active and to
be 0 otherwise. (It is always an IndexedConstraint because each
bound becomes a separate constraint.)
Parameters
----------
v: a Var which was created by the hull transformation as a
disaggregated variable (and so appears on a transformation
block of some Disjunct)
"""
# This can only go well if v is a disaggregated var
transBlock = v.parent_block().parent_block()
try:
return transBlock._bigMConstraintMap[v]
except:
logger.error("Either '%s' is not a disaggregated variable, or "
"the disjunction that disaggregates it has not "
"been properly transformed." % v.name)
raise