diff options
author | Nao Pross <np@0hm.ch> | 2024-06-05 18:01:27 +0200 |
---|---|---|
committer | Nao Pross <np@0hm.ch> | 2024-06-05 18:01:27 +0200 |
commit | e6a40dcfb2a4f837366332c7ceeacaba7d3f8c24 (patch) | |
tree | 64345bf60f497749c52d42f41acc48c5bb8fef83 | |
parent | Remove concatenation of constraints in SOSProblem.apply (diff) | |
download | sumofsquares-e6a40dcfb2a4f837366332c7ceeacaba7d3f8c24.tar.gz sumofsquares-e6a40dcfb2a4f837366332c7ceeacaba7d3f8c24.zip |
Fix bugs in PutinarPSatz, better names (variables -> symbols)
-rw-r--r-- | sumofsquares/canon.py | 95 | ||||
-rw-r--r-- | sumofsquares/problems.py | 41 |
2 files changed, 92 insertions, 44 deletions
diff --git a/sumofsquares/canon.py b/sumofsquares/canon.py index 8c4b541..9ba0639 100644 --- a/sumofsquares/canon.py +++ b/sumofsquares/canon.py @@ -9,11 +9,12 @@ from typing_extensions import override from polymatrix.expression.from_ import Expression from polymatrix.expression.init import init_variable_expr from polymatrix.expressionstate import ExpressionState +from polymatrix.polymatrix.index import VariableIndex from .abc import Problem, Solver, Constraint, Result from .constraints import NonNegative, PositiveSemiDefinite, ExponentialCone from .problems import SOSProblem, InternalSOSProblem -from .variable import from_name as opt_variable_from_name +from .variable import OptSymbol, from_name as opt_variable_from_name from .utils import partition @@ -26,7 +27,7 @@ class Canonicalization(Problem): s = f"Apply canonicalization procedure {self.__class__.__qualname__} to\n" return s + str(self.problem) - def __call__(self, problem : SOSProblem | None = None) -> SOSProblem: + def __call__(self, problem : SOSProblem) -> SOSProblem: # TODO: is this a good idea? probably not but I don't want to introduce # a new type of class for "horizontal" function in the commutative # diagram. The idea is to make canon objects behave like functions @@ -35,8 +36,7 @@ class Canonicalization(Problem): # computation, which would need a statemonad, defeating the purpose of # creating this class in the first place. """ - Apply the canonicalization procedure to the given problem, or to - `self.problem` if `problem` is None. + Apply the canonicalization procedure to the given problem. Override this method to write transformation if it can be written without having to apply the state. I.e. it is not stateful in the @@ -59,8 +59,7 @@ class Canonicalization(Problem): @property @override def solver(self) -> Solver: - # TODO: error message - raise AttributeError + return self.problem.solver @property @abstractmethod @@ -92,6 +91,12 @@ class PutinarPSatz(Canonicalization): problem: SOSProblem + @property + @override + def cost(self) -> Expression: + # PSatz does not change cost function + return self.problem.cost + @override def apply(self, state: ExpressionState) -> tuple[ExpressionState, InternalSOSProblem]: @@ -102,18 +107,57 @@ class PutinarPSatz(Canonicalization): constraints = tuple(constraints) to_process = tuple(to_process) - # Evaluate expression to gather all variables + # Variables that are involved in the problem + variable_indices: set[VariableIndex] = set() + + # Evaluate expressions of constraints tha need to be processed + # to gather all variables for c in to_process: - state, _ = c.expression.apply(state) + state, pm = c.expression.apply(state) + variable_indices |= pm.variables() + for p in c.domain.polynomials: - state, _ = p.apply(state) + state, pm = p.apply(state) + variable_indices |= pm.variables() + + # Get symbols of the variables + to_process_symbols = set(state.get_symbol_from_variable_index(v) + for v in variable_indices) + + # Separate between symbols of optimization variables and those of polynomial variables + def is_opt(v): + return isinstance(v, OptSymbol) + polynomial_symbols, symbols = partition(is_opt, to_process_symbols) + + # Get variables from rest of the problem state, prob = replace(self.problem, constraints=constraints).apply(state) + symbols = tuple(set(symbols) | set(prob.symbols)) + polynomial_symbols = tuple(set(polynomial_symbols) | set(prob.polynomial_symbols)) + + # Vector of polynomial variables + x = poly.v_stack((1,) + tuple(init_variable_expr(v, state.get_shape(v)) + for v in polynomial_symbols)) # Create new constraints new_constraints: list[Constraint] = list(constraints) for i, constr in enumerate(to_process): - new_constr = constr.expression + # FIXME: create an extension poly_degree and opt_degree to get + # the maximul degree of the polynomial and the degree of the + # optimization variables respectively. + constr_deg = constr.expression.linear_in(x).degree().max() + + # Add first term in the quadratic module + d = constr_deg + if d.read(state).scalar().constant() % 2 != 0: + d += 1 + + monomials = x.combinations(poly.arange(d + 1)).half_newton_polytope(x) + coeffs = opt_variable_from_name(rf"\gamma_{i},c", shape=monomials.shape) + + new_constr = constr.expression - coeffs.T @ monomials + + # Add other terms of the quadratic module for j, domain_poly in enumerate(constr.domain.polynomials): # Make a multiplier polynomial for a domain polynomial of a constraint. # @@ -123,11 +167,7 @@ class PutinarPSatz(Canonicalization): # :math:`p` is the domain polynomial and :math:`c` is the contraint # polynomial. - # FIXME: create an extension poly_degree and opt_degree to get - # the maximul degree of the polynomial and the degree of the - # optimization variables respectively. - constr_deg = constr.expression.eval({v: 1. for v in prob.variables}).degree() - domain_deg = domain_poly.eval({v: 1. for v in prob.variables}).degree() + domain_deg = domain_poly.linear_in(x).degree().max() # Degree of multiplier needs to be an integer d = constr_deg - domain_deg @@ -140,19 +180,16 @@ class PutinarPSatz(Canonicalization): if d.read(state).scalar().constant() % 2 != 0: d += 1 - x = poly.v_stack((1,) + tuple(init_variable_expr(v, state.get_shape(v)) - for v in prob.polynomial_variables)) - # FIXME: need to check that there is not a \gamma_i,j already! # TODO: deterministically generate unique names # TODO: better documentation # Generate monomials of up to degree d and take only monomials # that are in half of the newton polytope - monomials = x.combinations(poly.arange(d + 1)) # .half_newton_polytope(x) - gamma = opt_variable_from_name(rf"\gamma_{i},{j}", shape=monomials.shape) + monomials = x.combinations(poly.arange(d + 1)).half_newton_polytope(x) + coeffs = opt_variable_from_name(rf"\gamma_{i},{j}", shape=monomials.shape) - multiplier = gamma.T @ monomials + multiplier = coeffs.T @ monomials # Subtract multiplier from expression and impose that it is also SOS new_constr -= multiplier * domain_poly @@ -211,11 +248,18 @@ class LogDet(Canonicalization): """ problem: SOSProblem + @property + @override + def cost(self) -> Expression: + return self(self.problem).cost + + @property @override - def __call__(self, problem : SOSProblem | None = None) -> SOSProblem: - if problem is None: - problem = self.problem + def constraints(self) -> Expression: + return self(self.problem).constraints + @override + def __call__(self, problem : SOSProblem) -> SOSProblem: A = problem.cost # FIXME: should check for name clashes in state object @@ -236,7 +280,6 @@ class LogDet(Canonicalization): return replace(problem, cost=new_cost, constraints=new_constraints) - @override def apply(self, state: ExpressionState) -> tuple[ExpressionState, InternalSOSProblem]: - return self().apply(state) + return self(self.problem).apply(state) diff --git a/sumofsquares/problems.py b/sumofsquares/problems.py index a646d0e..9990264 100644 --- a/sumofsquares/problems.py +++ b/sumofsquares/problems.py @@ -179,31 +179,31 @@ class SOSProblem(Problem): variable_indices: set[VariableIndex] = set() constraints: list[Constraint] = [] + # Collect all variables that are involved in the problem state, cost = self.cost.apply(state) - variable_indices.update(cost.variables()) - - # Compute the polymatrix of each constraint expression. Even though the - # result may not be used, this is necessary to account for the case of - # variables that are only present in the constraints, which the state - # object may not contain yet. + variable_indices |= cost.variables() for c in self.constraints: state, pm = c.expression.apply(state) - variable_indices.update(pm.variables()) + variable_indices |= pm.variables() - variables = set(state.get_symbol_from_variable_index(v) - for v in variable_indices) + # Get symbols of the variables + all_symbols = set(state.get_symbol_from_variable_index(v) + for v in variable_indices) - # Collect optimization variables + # Separate between symbols of optimization variables and those of polynomial variables def is_opt(v): return isinstance(v, OptSymbol) - polynomial_variables, variables = partition(is_opt, variables) - polynomial_variables = tuple(polynomial_variables) # because it is a generator + polynomial_symbols, symbols = partition(is_opt, all_symbols) + + # evaluate generators + symbols = tuple(symbols) + polynomial_symbols = tuple(polynomial_symbols) x = poly.v_stack((1,) + tuple( init_variable_expr(v, state.get_shape(v)) - for v in polynomial_variables)) + for v in polynomial_symbols)) for i, c in enumerate(self.constraints): if isinstance(c, EqualToZero): @@ -287,7 +287,7 @@ class SOSProblem(Problem): # raise NotImplementedError(f"Cannot process constraint of type {ctype} (yet).") return state, InternalSOSProblem(cost, tuple(pm_constraints), - tuple(variables), polynomial_variables, + symbols, polynomial_symbols, self.solver) @@ -305,8 +305,8 @@ class InternalSOSProblem(Problem): """ cost: PolyMatrixMixin constraints: Sequence[Constraint[PolyMatrixMixin]] - variables: Sequence[OptSymbol] - polynomial_variables: Sequence[Symbol] + symbols: Sequence[OptSymbol] + polynomial_symbols: Sequence[Symbol] solver: Solver def to_conic_problem(self, state: ExpressionState, verbose: bool = False) -> ConicProblem: @@ -335,7 +335,7 @@ class InternalSOSProblem(Problem): # indices of variables in the optimization problem, sorted variable_indices = sum(sorted(tuple(state.get_indices_as_variable_index(v)) - for v in self.variables), ()) + for v in self.symbols), ()) # cost linear term cost_coeffs = dict(cost.affine_coefficients_by_degrees(variable_indices)) @@ -374,6 +374,11 @@ class InternalSOSProblem(Problem): # Get constant and linear terms constant = constr.affine_coefficient(MonomialIndex.constant()) linear = tuple(constr.affine_coefficient(v) for v in variable_indices) + + # constraint has disappeared + if not (np.any(constant) or np.any(linear)): + # print("Constraint disappeared!") + continue if isinstance(c, EqualToZero): dims["z"].append(nrows) @@ -420,7 +425,7 @@ class InternalSOSProblem(Problem): dims=dims, is_qp=is_qp, solver=self.solver, variables={v : state.get_shape(v) - for v in self.variables + for v in self.symbols }) |