def build_sparse_linear_model_and_solve(cls, nb_vars, var_lbs, var_ubs,
var_types, var_names,
nb_rows, cts_sparse_coefs,
objsense, costs,
solution_maker=make_solution,
**transform_params):
cpx = cls.create_cplex()
# varlist = mdl.continuous_var_list(var_count, lb=var_lbs, ub=var_ubs, name=var_names)
cls.create_column_vars(cpx, nb_vars, var_lbs, var_ubs, var_types, var_names)
var_indices = list(range(nb_vars))
cpx_linexprs = [([], []) for _ in range(nb_rows)]
cpx_rhss = [0.] * nb_rows
for coef, row, col in cts_sparse_coefs:
if col >= nb_vars:
cpx_rhss[row] = float(coef)
elif coef:
cpx_row = cpx_linexprs[row]
# int() conversio nis mandatory here
# as sparse matrices contain numpy int types -> cause cplex to crash
cpx_row[0].append(int(col))
cpx_row[1].append(float(coef))
ctsense = ComparisonType.parse(transform_params.get('sense', 'le'))
cpx_senses = ctsense.cplex_code * nb_rows
#fast_add_linear(cpx, cpx_linexprs, cpx_senses, cpx_rhss, names=[])
cpx.linear_constraints.add(cpx_linexprs, cpx_senses, cpx_rhss, names=[])
if costs is not None:
# set linear objective for all variables.
fcosts = [float(k) for k in costs]
static_fast_set_linear_obj(cpx, var_indices, fcosts)
cpx.objective.set_sense(objsense.cplex_coef)
# here we go to solve...
return cls._solve(cpx, var_names, solution_maker=solution_maker, **transform_params)
def set_linear_constraint_sense(self, ct, arg_newsense):
new_sense = ComparisonType.parse(arg_newsense)
if new_sense != ct.sense:
self._engine.update_constraint(ct,
UpdateEvent.LinearConstraintType,
new_sense)
ct._internal_set_sense(new_sense)
def set_quadratic_constraint_sense(self, qct, arg_newsense):
new_sense = ComparisonType.parse(arg_newsense)
if new_sense != qct.sense:
self._engine.update_constraint(qct,
UpdateEvent.LinearConstraintType,
new_sense)
qct._internal_set_sense(new_sense)
def is_satisfied(self, solution, tolerance=1e-6):
left_value = self._left_expr._get_solution_value(solution)
right_value = self._right_expr._get_solution_value(solution)
return ComparisonType.almost_compare(left_value,
self._ctsense,
right_value,
eps=tolerance)
def vector_compare(self, lhss, rhss, sense):
l_lhs = self._to_list(lhss, caller='Model.vector.compare')
l_rhs = self._to_list(rhss, caller='Model.vector.compare')
if len(l_lhs) != len(l_rhs):
self.fatal(
'Model.vector_compare() got sequences with different length, left: {0}, right: {1}'
.format(len(l_lhs), len(l_rhs)))
ctsense = ComparisonType.parse(sense)
return self._aggregator.vector_compare(l_lhs, l_rhs, ctsense)
def build_matrix_linear_model_and_solve(cls,
var_count,
var_lbs,
var_ubs,
var_types,
var_names,
cts_mat,
rhs,
objsense,
costs,
cast_to_float,
solution_maker=make_solution,
**transform_params):
adapter = cls.create_cplex_adapter()
cpx = adapter.cpx
if cast_to_float:
print("-- all numbers will be cast to float")
else:
print("-- no cast to float is performed")
cls.create_column_vars(cpx, var_count, var_lbs, var_ubs, var_types,
var_names)
var_indices = list(range(var_count))
gen_rows = ModelAggregator.generate_rows(cts_mat)
cpx_rows = []
if cast_to_float:
for row in gen_rows:
# need this step as cplex may crash with np types.
frow = [float(k) for k in row]
cpx_rows.append([var_indices, frow])
else:
cpx_rows = [[var_indices, row] for row in gen_rows]
nb_rows = len(cpx_rows)
if nb_rows:
ctsense = ComparisonType.parse(transform_params.get('sense', 'le'))
cpx_senses = ctsense.cplex_code * nb_rows
cpx_rhss = [float(r) for r in rhs] if cast_to_float else rhs
adapter.add_linear(cpx, cpx_rows, cpx_senses, cpx_rhss, names=[])
if costs is not None:
# set linear objective for all variables.
fcosts = [float(k) for k in costs]
adapter.static_fast_set_linear_obj(cpx, var_indices, fcosts)
cpx.objective.set_sense(objsense.cplex_coef)
# here we go to solve...
return cls._solve(cpx,
var_names,
solution_maker=solution_maker,
**transform_params)
def as_constraint_from_symbol(self, op_symbol):
self_var = self.var
var_lb = self.var.lb
op = ComparisonType.cplex_ctsense_to_python_op(op_symbol)
ct = op(self_var, var_lb)
return ct
def new_binary_constraint(self, lhs, sense, rhs, name=None):
ctsense = ComparisonType.parse(sense)
return self._new_binary_constraint(lhs, ctsense, rhs, name)
def read(cls,
filename,
model_name=None,
verbose=False,
model_class=None,
**kwargs):
""" Reads a model from a CPLEX export file.
Accepts all formats exported by CPLEX: LP, SAV, MPS.
If an error occurs while reading the file, the message of the exception
is printed and the function returns None.
Args:
filename: The file to read.
model_name: An optional name for the newly created model. If None,
the model name will be the path basename.
verbose: An optional flag to print informative messages, default is False.
model_class: An optional class type; must be a subclass of Model.
The returned model is built using this model_class and the keyword arguments kwargs, if any.
By default, the model is class is `Model` (see
kwargs: A dict of keyword-based arguments that are used when creating the model
instance.
Example:
`m = read_model("c:/temp/foo.mps", model_name="docplex_foo", solver_agent="docloud", output_level=100)`
Returns:
An instance of Model, or None if an exception is raised.
See Also:
:class:`docplex.mp.model.Model`
"""
if not os.path.exists(filename):
raise IOError("* file not found: {0}".format(filename))
# extract basename
if model_name:
name_to_use = model_name
else:
basename = os.path.basename(filename)
if '.' not in filename:
raise RuntimeError(
'ModelReader.read_model(): path has no extension: {}'.
format(filename))
dotpos = basename.find(".")
if dotpos > 0:
name_to_use = basename[:dotpos]
else: # pragma: no cover
name_to_use = basename
model_class = model_class or Model
if 0 == os.stat(filename).st_size:
print("* file is empty: {0} - exiting".format(filename))
return model_class(name=name_to_use, **kwargs)
if verbose:
print("-> CPLEX starts reading file: {0}".format(filename))
cpx_adapter = cls._read_cplex(filename)
cpx = cpx_adapter.cpx
if verbose:
print("<- CPLEX finished reading file: {0}".format(filename))
if not cpx: # pragma: no cover
return None
final_output_level = kwargs.get("output_level", "info")
debug_read = kwargs.get("debug", False)
try:
# force no tck
if 'checker' in kwargs:
final_checker = kwargs['checker']
else:
final_checker = 'default'
# build the model with no checker, then restore final_checker in the end.
kwargs['checker'] = 'off'
ignore_names = kwargs.get('ignore_names', False)
# -------------
mdl = model_class(name=name_to_use, **kwargs)
lfactory = mdl._lfactory
qfactory = mdl._qfactory
mdl.set_quiet() # output level set to ERROR
vartype_cont = mdl.continuous_vartype
vartype_map = {
'B': mdl.binary_vartype,
'I': mdl.integer_vartype,
'C': mdl.continuous_vartype,
'S': mdl.semicontinuous_vartype
}
# 1 upload variables
cpx_nb_vars = cpx.variables.get_num()
def make_constant_expr(k):
if k:
return lfactory._new_safe_constant_expr(k)
else:
return lfactory.new_zero_expr()
if verbose:
print("-- uploading {0} variables...".format(cpx_nb_vars))
cpx_var_names = [] if ignore_names else cls._safe_call_get_names(
cpx_adapter, cpx.variables.get_names)
if cpx._is_MIP():
cpx_vartypes = [
vartype_map.get(cpxt, vartype_cont)
for cpxt in cpx.variables.get_types()
]
else:
cpx_vartypes = [vartype_cont] * cpx_nb_vars
cpx_var_lbs = cpx.variables.get_lower_bounds()
cpx_var_ubs = cpx.variables.get_upper_bounds()
# map from cplex variable indices to docplex's
# use to skip range vars
# cplex : [x, Rg1, y] -> {0:0, 2: 1}
if cpx_var_names:
model_varnames = cpx_var_names
else:
model_varnames = [None] * cpx_nb_vars
model_lbs = cpx_var_lbs
model_ubs = cpx_var_ubs
model_types = cpx_vartypes
# vars
model_vars = lfactory.new_multitype_var_list(
cpx_nb_vars, model_types, model_lbs, model_ubs, model_varnames)
# inverse map from indices to docplex vars
cpx_var_index_to_docplex = {
v: model_vars[v]
for v in range(cpx_nb_vars)
}
# 2. upload linear constraints and ranges (mixed in cplex)
cpx_linearcts = cpx.linear_constraints
nb_linear_cts = cpx_linearcts.get_num()
# all_rows1 = cpx_linearcts.get_rows()
all_rows = cpx_adapter.fast_get_rows(cpx)
all_rhs = cpx_linearcts.get_rhs()
all_senses = cpx_linearcts.get_senses()
all_range_values = cpx_linearcts.get_range_values()
cpx_ctnames = [] if ignore_names else cls._safe_call_get_names(
cpx_adapter, cpx_linearcts.get_names)
deferred_cts = []
if verbose:
print("-- uploading {0} linear constraints...".format(
nb_linear_cts))
for c in range(nb_linear_cts):
row = all_rows[c]
sense = all_senses[c]
rhs = all_rhs[c]
ctname = cpx_ctnames[c] if cpx_ctnames else None
range_val = all_range_values[c]
indices, coefs = row
expr = cls._make_expr_from_varmap_coefs(
lfactory, cpx_var_index_to_docplex, indices, coefs)
if sense == 'R':
# rangeval can be negative !!! issue 52
if range_val >= 0:
range_lb = rhs
range_ub = rhs + range_val
else:
range_ub = rhs
range_lb = rhs + range_val
rgct = mdl.range_constraint(lb=range_lb,
ub=range_ub,
expr=expr,
rng_name=ctname)
deferred_cts.append(rgct)
else:
op = cls.parse_sense(sense)
rhs_expr = make_constant_expr(rhs)
ct = LinearConstraint(mdl, expr, op, rhs_expr, ctname)
deferred_cts.append(ct)
if deferred_cts:
# add constraint as a block
lfactory._post_constraint_block(posted_cts=deferred_cts)
# 3. upload Quadratic constraints
cpx_quadraticcts = cpx.quadratic_constraints
nb_quadratic_cts = cpx_quadraticcts.get_num()
if nb_quadratic_cts:
all_rhs = cpx_quadraticcts.get_rhs()
all_linear_nb_non_zeros = cpx_quadraticcts.get_linear_num_nonzeros(
)
all_linear_components = cpx_quadraticcts.get_linear_components(
)
all_quadratic_nb_non_zeros = cpx_quadraticcts.get_quad_num_nonzeros(
)
all_quadratic_components = cpx_quadraticcts.get_quadratic_components(
)
all_senses = cpx_quadraticcts.get_senses()
cpx_ctnames = [] if ignore_names else cls._safe_call_get_names(
cpx_adapter, cpx_quadraticcts.get_names)
for c in range(nb_quadratic_cts):
rhs = all_rhs[c]
linear_nb_non_zeros = all_linear_nb_non_zeros[c]
linear_component = all_linear_components[c]
quadratic_nb_non_zeros = all_quadratic_nb_non_zeros[c]
quadratic_component = all_quadratic_components[c]
sense = all_senses[c]
ctname = cpx_ctnames[c] if cpx_ctnames else None
if linear_nb_non_zeros > 0:
indices, coefs = linear_component.unpack()
# linexpr = mdl._aggregator._scal_prod((cpx_var_index_to_docplex[idx] for idx in indices), coefs)
linexpr = cls._make_expr_from_varmap_coefs(
lfactory, cpx_var_index_to_docplex, indices, coefs)
else:
linexpr = None
if quadratic_nb_non_zeros > 0:
qfactory = mdl._qfactory
ind1, ind2, coefs = quadratic_component.unpack()
quads = qfactory.term_dict_type()
for idx1, idx2, coef in izip(ind1, ind2, coefs):
quads[VarPair(
cpx_var_index_to_docplex[idx1],
cpx_var_index_to_docplex[idx2])] = coef
else: # pragma: no cover
# should not happen, but who knows
quads = None
quad_expr = mdl._aggregator._quad_factory.new_quad(
quads=quads, linexpr=linexpr, safe=True)
op = ComparisonType.cplex_ctsense_to_python_op(sense)
ct = op(quad_expr, rhs)
mdl.add_constraint(ct, ctname)
# 4. upload indicators
cpx_indicators = cpx.indicator_constraints
nb_indicators = cpx_indicators.get_num()
if nb_indicators:
all_ind_names = [] if ignore_names else cls._safe_call_get_names(
cpx_adapter, cpx_indicators.get_names)
all_ind_bvars = cpx_indicators.get_indicator_variables()
all_ind_rhs = cpx_indicators.get_rhs()
all_ind_linearcts = cpx_indicators.get_linear_components()
all_ind_senses = cpx_indicators.get_senses()
all_ind_complemented = cpx_indicators.get_complemented()
all_ind_types = cpx_indicators.get_types()
ind_equiv_type = 3
for i in range(nb_indicators):
ind_bvar = all_ind_bvars[i]
ind_name = all_ind_names[i] if all_ind_names else None
ind_rhs = all_ind_rhs[i]
ind_linear = all_ind_linearcts[i] # SparsePair(ind, val)
ind_sense = all_ind_senses[i]
ind_complemented = all_ind_complemented[i]
ind_type = all_ind_types[i]
# 1 . check the bvar is ok
ind_bvar = cpx_var_index_to_docplex[ind_bvar]
# each var appears once
ind_linexpr = cls._build_linear_expr_from_sparse_pair(
lfactory, cpx_var_index_to_docplex, ind_linear)
op = ComparisonType.cplex_ctsense_to_python_op(ind_sense)
ind_lct = op(ind_linexpr, ind_rhs)
if ind_type == ind_equiv_type:
logct = lfactory.new_equivalence_constraint(
ind_bvar,
ind_lct,
true_value=1 - ind_complemented,
name=ind_name)
else:
logct = lfactory.new_indicator_constraint(
ind_bvar,
ind_lct,
true_value=1 - ind_complemented,
name=ind_name)
mdl.add(logct)
# 5. upload Piecewise linear constraints
try:
cpx_pwl = cpx.pwl_constraints
cpx_pwl_defs = cpx_pwl.get_definitions()
pwl_fallback_names = [""] * cpx_pwl.get_num()
cpx_pwl_names = pwl_fallback_names if ignore_names else cls._safe_call_get_names(
cpx_adapter, cpx_pwl.get_names, pwl_fallback_names)
for (vary_idx, varx_idx, preslope, postslope, breakx,
breaky), pwl_name in izip(cpx_pwl_defs, cpx_pwl_names):
varx = cpx_var_index_to_docplex.get(varx_idx, None)
vary = cpx_var_index_to_docplex.get(vary_idx, None)
breakxy = [(brkx, brky)
for brkx, brky in zip(breakx, breaky)]
pwl_func = mdl.piecewise(preslope,
breakxy,
postslope,
name=pwl_name)
pwl_expr = mdl._lfactory.new_pwl_expr(
pwl_func,
varx,
0,
add_counter_suffix=False,
resolve=False)
pwl_expr._f_var = vary
pwl_expr._ensure_resolved()
except AttributeError: # pragma: no cover
pass # Do not check for PWLs if Cplex version does not support them
# 6. upload objective
# noinspection PyPep8
try:
cpx_multiobj = cpx.multiobj
except AttributeError: # pragma: no cover
# pre-12.9 version
cpx_multiobj = None
if cpx_multiobj is None or cpx_multiobj.get_num() <= 1:
cpx_obj = cpx.objective
cpx_sense = cpx_obj.get_sense()
cpx_all_lin_obj_coeffs = cpx_obj.get_linear()
all_obj_vars = []
all_obj_coefs = []
for v in range(cpx_nb_vars):
if v in cpx_var_index_to_docplex:
obj_coeff = cpx_all_lin_obj_coeffs[v]
all_obj_coefs.append(obj_coeff)
all_obj_vars.append(cpx_var_index_to_docplex[v])
# obj_expr = mdl._aggregator._scal_prod(all_obj_vars, all_obj_coefs)
obj_expr = cls._make_expr_from_vars_coefs(
mdl, all_obj_vars, all_obj_coefs)
if cpx_obj.get_num_quadratic_variables() > 0:
cpx_all_quad_cols_coeffs = cpx_obj.get_quadratic()
quads = qfactory.term_dict_type()
for v, col_coefs in izip(cpx_var_index_to_docplex,
cpx_all_quad_cols_coeffs):
var1 = cpx_var_index_to_docplex[v]
indices, coefs = col_coefs.unpack()
for idx, coef in izip(indices, coefs):
vp = VarPair(var1, cpx_var_index_to_docplex[idx])
quads[vp] = quads.get(vp, 0) + coef / 2
obj_expr += qfactory.new_quad(quads=quads, linexpr=None)
obj_expr += cpx.objective.get_offset()
is_maximize = cpx_sense == cpx_adapter.cplex_module._internal._subinterfaces.ObjSense.maximize
if is_maximize:
mdl.maximize(obj_expr)
else:
mdl.minimize(obj_expr)
else:
# we have multiple objective
nb_multiobjs = cpx_multiobj.get_num()
exprs = [0] * nb_multiobjs
priorities = [1] * nb_multiobjs
weights = [1] * nb_multiobjs
abstols = [0] * nb_multiobjs
reltols = [0] * nb_multiobjs
names = cpx_multiobj.get_names()
for m in range(nb_multiobjs):
(obj_coeffs, obj_offset, weight, prio, abstol,
reltol) = cpx_multiobj.get_definition(m)
obj_expr = cls._make_expr_from_coef_vector(
mdl, cpx_var_index_to_docplex, obj_coeffs, obj_offset)
exprs[m] = obj_expr
priorities[m] = prio
weights[m] = weight
abstols[m] = abstol
reltols[m] = reltol
sense = cpx_multiobj.get_sense()
mdl.set_multi_objective(sense, exprs, priorities, weights,
abstols, reltols, names)
# upload sos
cpx_sos = cpx.SOS
cpx_sos_num = cpx_sos.get_num()
if cpx_sos_num > 0:
cpx_sos_types = cpx_sos.get_types()
cpx_sos_indices = cpx_sos.get_sets()
cpx_sos_names = cpx_sos.get_names()
if not cpx_sos_names:
cpx_sos_names = [None] * cpx_sos_num
for sostype, sos_sparse, sos_name in izip(
cpx_sos_types, cpx_sos_indices, cpx_sos_names):
sos_var_indices = sos_sparse.ind
sos_weights = sos_sparse.val
isostype = int(sostype)
sos_vars = [
cpx_var_index_to_docplex[var_ix]
for var_ix in sos_var_indices
]
mdl.add_sos(dvars=sos_vars,
sos_arg=isostype,
name=sos_name,
weights=sos_weights)
# upload lazy constraints
cpx_linear_advanced = cpx.linear_constraints.advanced
cpx_lazyct_num = cpx_linear_advanced.get_num_lazy_constraints()
if cpx_lazyct_num:
print(
"WARNING: found {0} lazy constraints that cannot be uploaded to DOcplex"
.format(cpx_lazyct_num))
mdl.output_level = final_output_level
if final_checker:
# need to restore checker
mdl.set_checker(final_checker)
except cpx_adapter.CplexError as cpx_e: # pragma: no cover
print("* CPLEX error: {0!s} reading file {1}".format(
cpx_e, filename))
mdl = None
if debug_read:
raise
except ModelReaderError as mre: # pragma: no cover
print("! Model reader error: {0!s} while reading file {1}".format(
mre, filename))
mdl = None
if debug_read:
raise
except DOcplexException as doe: # pragma: no cover
print("! Internal DOcplex error: {0!s} while reading file {1}".
format(doe, filename))
mdl = None
if debug_read:
raise
# except Exception as any_e: # pragma: no cover
# print("Internal exception raised: {0} msg={1!s} while reading file '{2}'".format(type(any_e), any_e, filename))
# mdl = None
# if debug_read:
# raise
finally:
# clean up CPLEX instance...
cpx.end()
return mdl
def read_model(self,
filename,
model_name=None,
verbose=False,
model_class=None,
**kwargs):
""" Reads a model from a CPLEX export file.
Accepts all formats exported by CPLEX: LP, SAV, MPS.
If an error occurs while reading the file, the message of the exception
is printed and the function returns None.
Args:
filename: The file to read.
model_name: An optional name for the newly created model. If None,
the model name will be the path basename.
verbose: An optional flag to print informative messages, default is False.
model_class: An optional class type; must be a subclass of Model.
The returned model is built using this model_class and the keyword arguments kwargs, if any.
By default, the model is class is `Model` (see
kwargs: A dict of keyword-based arguments that are used when creating the model
instance.
Example:
`m = read_model("c:/temp/foo.mps", model_name="docplex_foo", solver_agent="docloud", output_level=100)`
Returns:
An instance of Model, or None if an exception is raised.
See Also:
:class:`docplex.mp.model.Model`
"""
if not Cplex: # pragma: no cover
raise RuntimeError(
"ModelReader.read_model() requires CPLEX runtime.")
if not os.path.exists(filename):
raise IOError("* file not found: {0}".format(filename))
# extract basename
if model_name:
name_to_use = model_name
else:
basename = os.path.basename(filename)
dotpos = basename.find(".")
if dotpos > 0:
name_to_use = basename[:dotpos]
else:
name_to_use = basename
model_class = model_class or Model
if 0 == os.stat(filename).st_size:
print("* file is empty: {0} - exiting".format(filename))
return model_class(name=name_to_use, **kwargs)
# print("-> start reading file: {0}".format(filename))
cpx = self._cplex_read(filename, verbose=verbose)
if not cpx: # pragma: no cover
return None
range_map = {}
final_output_level = kwargs.get("output_level", "info")
debug_read = kwargs.get("debug", False)
try:
# force no tck
if 'checker' in kwargs:
final_checker = kwargs['checker']
else:
final_checker = 'default'
# build the model with no checker, then restore final_checker in the end.
kwargs['checker'] = 'off'
# -------------
mdl = model_class(name=name_to_use, **kwargs)
lfactory = mdl._lfactory
qfactory = mdl._qfactory
mdl.set_quiet() # output level set to ERROR
vartype_cont = mdl.continuous_vartype
vartype_map = {
'B': mdl.binary_vartype,
'I': mdl.integer_vartype,
'C': mdl.continuous_vartype,
'S': mdl.semicontinuous_vartype
}
# 1 upload variables
cpx_nb_vars = cpx.variables.get_num()
cpx_var_names = self._safe_call_get_names(cpx.variables)
if cpx._is_MIP():
cpx_vartypes = [
vartype_map.get(cpxt, vartype_cont)
for cpxt in cpx.variables.get_types()
]
else:
cpx_vartypes = [vartype_cont] * cpx_nb_vars
cpx_var_lbs = cpx.variables.get_lower_bounds()
cpx_var_ubs = cpx.variables.get_upper_bounds()
# map from cplex variable indices to docplex's
# use to skip range vars
# cplex : [x, Rg1, y] -> {0:0, 2: 1}
var_index_map = {}
d = 0
model_varnames = []
model_lbs = []
model_ubs = []
model_types = []
for v in range(cpx_nb_vars):
varname = cpx_var_names[v] if cpx_var_names else None
if varname and varname.startswith("Rg"):
# generated var for ranges
range_map[v] = self._RangeData(var_index=v,
var_name=varname,
ub=cpx_var_ubs[v])
else:
# docplex_var = lfactory.new_var(vartype, lb, ub, varname)
var_index_map[v] = d
model_varnames.append(varname)
model_types.append(cpx_vartypes[v])
model_lbs.append(cpx_var_lbs[v])
model_ubs.append(cpx_var_ubs[v])
d += 1
# vars
model_vars = lfactory.new_multitype_var_list(
d, model_types, model_lbs, model_ubs, model_varnames)
cpx_var_index_to_docplex = {
v: model_vars[var_index_map[v]]
for v in var_index_map.keys()
}
# 2. upload linear constraints and ranges (mixed in cplex)
cpx_linearcts = cpx.linear_constraints
nb_linear_cts = cpx_linearcts.get_num()
all_rows = cpx_linearcts.get_rows()
all_rhs = cpx_linearcts.get_rhs()
all_senses = cpx_linearcts.get_senses()
all_range_values = cpx_linearcts.get_range_values()
cpx_ctnames = self._safe_call_get_names(cpx_linearcts)
has_range = range_map or any(s == "R" for s in all_senses)
deferred_cts = []
for c in range(nb_linear_cts):
row = all_rows[c]
sense = all_senses[c]
rhs = all_rhs[c]
ctname = cpx_ctnames[c] if cpx_ctnames else None
range_val = all_range_values[c]
indices = row.ind
coefs = row.val
range_data = None
if not has_range:
expr = mdl._aggregator._scal_prod(
(cpx_var_index_to_docplex[idx] for idx in indices),
coefs)
op = ComparisonType.parse(sense)
ct = lfactory._new_binary_constraint(lhs=expr,
rhs=rhs,
sense=op)
ct.name = ctname
deferred_cts.append(ct)
else:
expr = lfactory.linear_expr()
rcoef = 1
for idx, koef in izip(indices, coefs):
var = cpx_var_index_to_docplex.get(idx, None)
if var:
expr._add_term(var, koef)
elif idx in range_map:
# this is a range: coeff must be 1 or -1
abscoef = koef if koef >= 0 else -koef
rcoef = koef
assert abscoef == 1, "range var has coef different from 1: {}".format(
koef)
assert range_data is None, "range_data is not None: {0!s}".format(
range_data) # cannot use two range vars
range_data = range_map[idx]
else: # pragma: no cover
# this is an internal error.
raise ModelReaderError(
"ERROR: index not in var map or range map: {0}"
.format(idx))
if range_data:
label = ctname or 'c#%d' % (c + 1)
if sense not in "EL": # pragma: no cover
raise ModelReaderError(
"{0} range sense is not E: {1!s}".format(
label, sense))
if rcoef < 0: # -1 actually
rng_lb = rhs
rng_ub = rhs + range_data.ub
elif rcoef > 0: # koef is 1 here
rng_lb = rhs - range_data.ub
rng_ub = rhs
else: # pragma: no cover
raise ModelReaderError(
"unexpected range coef: {}".format(rcoef))
mdl.add_range(lb=rng_lb,
expr=expr,
ub=rng_ub,
rng_name=ctname)
else:
if sense == 'R':
# range min is rangeval
range_lb = rhs
range_ub = rhs + range_val
mdl.add_range(lb=range_lb,
ub=range_ub,
expr=expr,
rng_name=ctname)
else:
op = ComparisonType.cplex_ctsense_to_python_op(
sense)
ct = op(expr, rhs)
mdl.add_constraint(ct, ctname)
if deferred_cts:
# add constraint as a block
lfactory._post_constraint_block(posted_cts=deferred_cts)
# 3. upload Quadratic constraints
cpx_quadraticcts = cpx.quadratic_constraints
nb_quadratic_cts = cpx_quadraticcts.get_num()
all_rhs = cpx_quadraticcts.get_rhs()
all_linear_nb_non_zeros = cpx_quadraticcts.get_linear_num_nonzeros(
)
all_linear_components = cpx_quadraticcts.get_linear_components()
all_quadratic_nb_non_zeros = cpx_quadraticcts.get_quad_num_nonzeros(
)
all_quadratic_components = cpx_quadraticcts.get_quadratic_components(
)
all_senses = cpx_quadraticcts.get_senses()
cpx_ctnames = self._safe_call_get_names(cpx_quadraticcts)
for c in range(nb_quadratic_cts):
rhs = all_rhs[c]
linear_nb_non_zeros = all_linear_nb_non_zeros[c]
linear_component = all_linear_components[c]
quadratic_nb_non_zeros = all_quadratic_nb_non_zeros[c]
quadratic_component = all_quadratic_components[c]
sense = all_senses[c]
ctname = cpx_ctnames[c] if cpx_ctnames else None
if linear_nb_non_zeros > 0:
indices, coefs = linear_component.unpack()
linexpr = mdl._aggregator._scal_prod(
(cpx_var_index_to_docplex[idx] for idx in indices),
coefs)
else:
linexpr = None
if quadratic_nb_non_zeros > 0:
qfactory = mdl._qfactory
ind1, ind2, coefs = quadratic_component.unpack()
quads = qfactory.term_dict_type()
for idx1, idx2, coef in izip(ind1, ind2, coefs):
quads[VarPair(cpx_var_index_to_docplex[idx1],
cpx_var_index_to_docplex[idx2])] = coef
else: # pragma: no cover
# should not happen, but who knows
quads = None
quad_expr = mdl._aggregator._quad_factory.new_quad(
quads=quads, linexpr=linexpr, safe=True)
op = ComparisonType.cplex_ctsense_to_python_op(sense)
ct = op(quad_expr, rhs)
mdl.add_constraint(ct, ctname)
# 4. upload indicators
cpx_indicators = cpx.indicator_constraints
nb_indicators = cpx_indicators.get_num()
all_ind_names = self._safe_call_get_names(cpx_indicators)
all_ind_bvars = cpx_indicators.get_indicator_variables()
all_ind_rhs = cpx_indicators.get_rhs()
all_ind_linearcts = cpx_indicators.get_linear_components()
all_ind_senses = cpx_indicators.get_senses()
all_ind_complemented = cpx_indicators.get_complemented()
lfactory = mdl._lfactory
for i in range(nb_indicators):
ind_bvar = all_ind_bvars[i]
ind_name = all_ind_names[i] if all_ind_names else None
ind_rhs = all_ind_rhs[i]
ind_linear = all_ind_linearcts[i] # SparsePair(ind, val)
ind_sense = all_ind_senses[i]
ind_complemented = all_ind_complemented[i]
# 1 . check the bvar is ok
ind_bvar = cpx_var_index_to_docplex[ind_bvar]
# each var appears once
ind_linexpr = self._build_linear_expr_from_sparse_pair(
lfactory, cpx_var_index_to_docplex, ind_linear)
op = ComparisonType.cplex_ctsense_to_python_op(ind_sense)
ind_ct = op(ind_linexpr, ind_rhs)
indct = lfactory.new_indicator_constraint(ind_bvar,
ind_ct,
active_value=1 -
ind_complemented,
name=ind_name)
mdl.add(indct)
# 5. upload Piecewise linear constraints
try:
cpx_pwl = cpx.pwl_constraints
cpx_pwl_defs = cpx_pwl.get_definitions()
pwl_fallback_names = [""] * cpx_pwl.get_num()
cpx_pwl_names = self._safe_call_get_names(
cpx_pwl, pwl_fallback_names)
for (vary_idx, varx_idx, preslope, postslope, breakx,
breaky), pwl_name in izip(cpx_pwl_defs, cpx_pwl_names):
varx = cpx_var_index_to_docplex.get(varx_idx, None)
vary = cpx_var_index_to_docplex.get(vary_idx, None)
breakxy = [(brkx, brky)
for brkx, brky in zip(breakx, breaky)]
pwl_func = mdl.piecewise(preslope,
breakxy,
postslope,
name=pwl_name)
pwl_expr = mdl._lfactory.new_pwl_expr(
pwl_func,
varx,
0,
add_counter_suffix=False,
resolve=False)
pwl_expr._f_var = vary
pwl_expr._ensure_resolved()
except AttributeError: # pragma: no cover
pass # Do not check for PWLs if Cplex version does not support them
# 6. upload objective
cpx_obj = cpx.objective
cpx_sense = cpx_obj.get_sense()
cpx_all_lin_obj_coeffs = cpx_obj.get_linear()
# noinspection PyPep8
all_obj_vars = []
all_obj_coefs = []
for v in range(cpx_nb_vars):
if v in cpx_var_index_to_docplex:
obj_coeff = cpx_all_lin_obj_coeffs[v]
all_obj_coefs.append(obj_coeff)
all_obj_vars.append(cpx_var_index_to_docplex[v])
# obj_expr._add_term(idx_to_var_map[v], cpx_all_obj_coeffs[v])
obj_expr = mdl._aggregator._scal_prod(all_obj_vars, all_obj_coefs)
if cpx_obj.get_num_quadratic_variables() > 0:
cpx_all_quad_cols_coeffs = cpx_obj.get_quadratic()
quads = qfactory.term_dict_type()
for v, col_coefs in izip(cpx_var_index_to_docplex,
cpx_all_quad_cols_coeffs):
var1 = cpx_var_index_to_docplex[v]
indices, coefs = col_coefs.unpack()
for idx, coef in izip(indices, coefs):
vp = VarPair(var1, cpx_var_index_to_docplex[idx])
quads[vp] = quads.get(vp, 0) + coef / 2
obj_expr += qfactory.new_quad(quads=quads, linexpr=None)
obj_expr += cpx.objective.get_offset()
is_maximize = cpx_sense == ObjSense.maximize
if is_maximize:
mdl.maximize(obj_expr)
else:
mdl.minimize(obj_expr)
# upload sos
cpx_sos = cpx.SOS
cpx_sos_num = cpx_sos.get_num()
if cpx_sos_num > 0:
cpx_sos_types = cpx_sos.get_types()
cpx_sos_indices = cpx_sos.get_sets()
cpx_sos_names = cpx_sos.get_names()
if not cpx_sos_names:
cpx_sos_names = [None] * cpx_sos_num
for sostype, sos_sparse, sos_name in izip(
cpx_sos_types, cpx_sos_indices, cpx_sos_names):
sos_var_indices = sos_sparse.ind
isostype = int(sostype)
sos_vars = [
cpx_var_index_to_docplex[var_ix]
for var_ix in sos_var_indices
]
mdl.add_sos(dvars=sos_vars,
sos_arg=isostype,
name=sos_name)
# upload lazy constraints
cpx_linear_advanced = cpx.linear_constraints.advanced
cpx_lazyct_num = cpx_linear_advanced.get_num_lazy_constraints()
if cpx_lazyct_num:
print(
"WARNING: found {0} lazy constraints that cannot be uploaded to DOcplex"
.format(cpx_lazyct_num))
mdl.output_level = final_output_level
if final_checker:
# need to restore checker
mdl.set_checker(final_checker)
except CplexError as cpx_e: # pragma: no cover
print("* CPLEX error: {0!s} reading file {1}".format(
cpx_e, filename))
mdl = None
if debug_read:
raise
except ModelReaderError as mre: # pragma: no cover
print("! Model reader error: {0!s} while reading file {1}".format(
mre, filename))
mdl = None
if debug_read:
raise
except DOcplexException as doe: # pragma: no cover
print("! Internal DOcplex error: {0!s} while reading file {1}".
format(doe, filename))
mdl = None
if debug_read:
raise
except Exception as any_e: # pragma: no cover
print("Internal exception raised: {0!s} while reading file {1}".
format(any_e, filename))
mdl = None
if debug_read:
raise
finally:
# clean up CPLEX instance...
del cpx
return mdl
def matrix_constraints(self, coef_mat, dvars, rhs, sense='le'):
"""
Creates a list of linear constraints
from a matrix of coefficients, a sequence of variables, and a sequence of numbers.
This method returns the list of constraints built from
A.X <op> B
where A is the coefficient matrix (of size (M,N)), X is the variable sequence (size N),
and B is the sequence of right-hand side values (of size M).
<op> is the comparison operator that defines the sense of the constraint. By default, this generates
a 'less-than-or-equal' constraint.
Example:
`Model.scal_prod_vars_triple([x, y], [z, t], [2, 3])` returns the expression `2xz + 3yt`.
:param coef_mat: A matrix of coefficients with M rows and N columns. This argument accepts
either a list of lists of numbers, a `numpy` array with size (M,N), or a `scipy` sparse matrix.
:param dvars: An ordered sequence of decision variables: accepts a Python list, `numpy` array,
or a `pandas` series. The size of the sequence must match the number of columns in the matrix.
:param rhs: A sequence of numbers: accepts a Python list, a `numpy` array,
or a `pandas` series. The size of the sequence must match the number of rows in the matrix.
:param sense: A constraint sense \; accepts either a
value of type `ComparisonType` or a string (e.g 'le', 'eq', 'ge').
:returns: A list of linear constraints.
Example:
If A is a matrix of coefficients with 2 rows and 3 columns::
A = [[1, 2, 3],
[4, 5, 6]],
X = [x, y, z] where x, y, and z are decision variables (size 3), and
B = [100, 200], a sequence of numbers (size 2),
then::
`mdl.matrix_constraint(A, X, B, 'GE')` returns a list of two constraints
[(x + 2y+3z <= 100), (4x + 5y +6z <= 200)].
Note:
If the dimensions of the matrix and variables or of the matrix and number sequence do not match,
an error is raised.
"""
checker = self._checker
if is_pandas_dataframe(coef_mat) or is_numpy_matrix(
coef_mat) or is_scipy_sparse(coef_mat):
nb_rows, nb_cols = coef_mat.shape
else:
# a sequence of sequences
a_mat = list(coef_mat)
nb_rows = len(a_mat)
nb_cols = None
try:
shared_len = None
for r in a_mat:
checker.check_ordered_sequence(r, 'matrix_constraints')
r_len = len(r)
if shared_len is None:
shared_len = r_len
elif r_len != shared_len:
self.fatal(
'All columns should have same length found {0} != {1}'
.format(shared_len, r_len))
nb_cols = shared_len if shared_len is not None else 0
except AttributeError:
self.fatal('All columns should have a len()')
s_dvars = self._to_list(dvars, caller='Model.matrix-constraints()')
s_rhs = self._to_list(rhs, caller='Model.matrix-constraints()')
# check
checker.typecheck_var_seq(s_dvars)
for k in s_rhs:
checker.typecheck_num(k)
op = ComparisonType.parse(sense)
# ---
# check dimensions and whether to transpose or not.
# ---
nb_rhs = len(s_rhs)
nb_vars = len(s_dvars)
if (nb_rows, nb_cols) != (nb_rhs, nb_vars):
self.fatal(
'Dimension error, matrix is ({0},{1}), expecting ({3}, {2})'.
format(nb_rows, nb_cols, nb_vars, nb_rhs))
if is_scipy_sparse(coef_mat):
return self._aggregator._sparse_matrix_constraints(
coef_mat, s_dvars, s_rhs, op)
else:
return self._aggregator._matrix_constraints(
coef_mat, s_dvars, s_rhs, op)
def get_constraint(self):
var_ub = self.get_ub()
op = ComparisonType.cplex_ctsense_to_python_op('L')
ct = op(self.get_var(), var_ub)
return ct
def is_satisfied(self, solution, tolerance):
expr_value = self._input_var._get_solution_value(solution)
y_value = solution._get_var_value(self._y)
computed_f_expr_value = self._pwl_func.evaluate(expr_value)
return ComparisonType.almost_equal(y_value, computed_f_expr_value, tolerance)
def is_satisfied(self, solution, tolerance=1e-6):
is_ct_satisfied = self._linear_ct.is_satisfied(solution, tolerance)
binary_value = solution.get_value(self._binary_var)
expected_value = self._active_value if is_ct_satisfied else 1 - self._active_value
return ComparisonType.almost_equal(binary_value, expected_value, tolerance)
