org.chocosolver.solver.constraints.nary.sum.PropSum Maven / Gradle / Ivy
Show all versions of choco-solver Show documentation
/*
* This file is part of choco-solver, http://choco-solver.org/
*
* Copyright (c) 2020, IMT Atlantique. All rights reserved.
*
* Licensed under the BSD 4-clause license.
*
* See LICENSE file in the project root for full license information.
*/
package org.chocosolver.solver.constraints.nary.sum;
import static org.chocosolver.solver.constraints.Operator.EQ;
import static org.chocosolver.solver.constraints.Operator.GE;
import static org.chocosolver.solver.constraints.Operator.LE;
import org.chocosolver.solver.constraints.Operator;
import org.chocosolver.solver.constraints.Propagator;
import org.chocosolver.solver.constraints.PropagatorPriority;
import org.chocosolver.solver.constraints.nary.clauses.ClauseBuilder;
import org.chocosolver.solver.exception.ContradictionException;
import org.chocosolver.solver.learn.ExplanationForSignedClause;
import org.chocosolver.solver.learn.Implications;
import org.chocosolver.solver.variables.IntVar;
import org.chocosolver.solver.variables.events.IntEventType;
import org.chocosolver.util.ESat;
import org.chocosolver.util.objects.ValueSortedMap;
import org.chocosolver.util.objects.setDataStructures.iterable.IntIterableRangeSet;
/**
* A propagator for SUM(x_i) o b
*
* Based on "Bounds Consistency Techniques for Long Linear Constraint"
* W. Harvey and J. Schimpf
*
*
* @author Charles Prud'homme
* @since 18/03/11
*/
public class PropSum extends Propagator {
/**
* The position of the last positive coefficient
*/
protected final int pos;
/**
* Number of variables
*/
protected final int l;
/**
* Bound to respect
*/
protected final int b;
/**
* Variability of each variable (ie domain amplitude)
*/
protected final int[] I;
/**
* Stores the maximal variability
*/
protected int maxI;
/**
* SUm of lower bounds
*/
protected int sumLB;
/**
* Sum of upper bounds
*/
protected int sumUB;
/**
* The operator among EQ, LE, GE and NE
*/
protected final Operator o;
/**
* Creates a sum propagator: SUM(x_i) o b
* Coefficients are induced by pos:
* those before pos (included) are equal to 1,
* the other ones are equal to -1.
*
* @param variables list of integer variables
* @param pos position of the last positive coefficient
* @param o operator amng EQ, LE, GE and NE
* @param b bound to respect
*/
public PropSum(IntVar[] variables, int pos, Operator o, int b) {
this(variables, pos, o, b, computePriority(variables.length), false);
}
PropSum(IntVar[] variables, int pos, Operator o, int b, PropagatorPriority priority, boolean reactOnFineEvent){
super(variables, priority, reactOnFineEvent);
this.pos = pos;
this.o = o;
this.b = b;
l = variables.length;
I = new int[l];
maxI = 0;
}
/**
* Compute the priority of the propagator wrt the number of involved variables
* @param nbvars number of variables
* @return the priority
*/
protected static PropagatorPriority computePriority(int nbvars) {
if (nbvars == 1) {
return PropagatorPriority.UNARY;
} else if (nbvars == 2) {
return PropagatorPriority.BINARY;
} else if (nbvars == 3) {
return PropagatorPriority.TERNARY;
} else {
return PropagatorPriority.LINEAR;
}
}
@Override
public int getPropagationConditions(int vIdx) {
switch (o) {
case NQ:
return IntEventType.instantiation();
case LE:
return IntEventType.combine(IntEventType.INSTANTIATE, vIdx < pos ? IntEventType.INCLOW : IntEventType.DECUPP);
case GE:
return IntEventType.combine(IntEventType.INSTANTIATE, vIdx < pos ? IntEventType.DECUPP : IntEventType.INCLOW);
default:
return IntEventType.boundAndInst();
}
}
/**
* Prepare the propagation: compute sumLB, sumUB and I
*/
protected void prepare() {
sumLB = sumUB = 0;
int i = 0;
int lb, ub;
maxI = 0;
for (; i < pos; i++) { // first the positive coefficients
lb = vars[i].getLB();
ub = vars[i].getUB();
sumLB += lb;
sumUB += ub;
I[i] = (ub - lb);
if(maxI < I[i])maxI = I[i];
}
for (; i < l; i++) { // then the negative ones
lb = -vars[i].getUB();
ub = -vars[i].getLB();
sumLB += lb;
sumUB += ub;
I[i] = (ub - lb);
if(maxI < I[i])maxI = I[i];
}
}
@Override
public void propagate(int evtmask) throws ContradictionException {
filter();
}
/**
* Execute filtering wrt the operator
* @throws ContradictionException if contradiction is detected
*/
protected void filter() throws ContradictionException {
prepare();
switch (o) {
case LE:
filterOnLeq();
break;
case GE:
filterOnGeq();
break;
case NQ:
filterOnNeq();
break;
default:
filterOnEq();
break;
}
}
/**
* Apply filtering when operator is EQ
* @throws ContradictionException if contradiction is detected
*/
protected void filterOnEq() throws ContradictionException {
boolean anychange;
int F = b - sumLB;
int E = sumUB - b;
do {
anychange = false;
// When explanations are on, no global failure allowed
if (model.getSolver().isLearnOff() && (F < 0 || E < 0)) {
fails();
}
if (maxI > F || maxI > E) {
int lb, ub, i = 0;
maxI = 0;
// positive coefficients first
while (i < pos) {
if (I[i] - F > 0) {
lb = vars[i].getLB();
ub = lb + I[i];
if (vars[i].updateUpperBound(F + lb, this)) {
int nub = vars[i].getUB();
E += nub - ub;
I[i] = nub - lb;
anychange = true;
}
}
if (I[i] - E > 0) {
ub = vars[i].getUB();
lb = ub - I[i];
if (vars[i].updateLowerBound(ub - E, this)) {
int nlb = vars[i].getLB();
F -= nlb - lb;
I[i] = ub - nlb;
anychange = true;
}
}
if(maxI < I[i])maxI = I[i];
i++;
}
// then negative ones
while (i < l) {
if (I[i] - F > 0) {
lb = -vars[i].getUB();
ub = lb + I[i];
if (vars[i].updateLowerBound(-F - lb, this)) {
int nub = -vars[i].getLB();
E += nub - ub;
I[i] = nub - lb;
anychange = true;
}
}
if (I[i] - E > 0) {
ub = -vars[i].getLB();
lb = ub - I[i];
if (vars[i].updateUpperBound(-ub + E, this)) {
int nlb = -vars[i].getUB();
F -= nlb - lb;
I[i] = ub - nlb;
anychange = true;
}
}
if(maxI < I[i])maxI = I[i];
i++;
}
}
// useless since true when all variables are instantiated
if (F <= 0 && E <= 0) {
this.setPassive();
return;
}
}while (anychange) ;
}
/**
* Apply filtering when operator is LE
* @throws ContradictionException if contradiction is detected
*/
protected void filterOnLeq() throws ContradictionException {
int F = b - sumLB;
int E = sumUB - b;
// When explanations are on, no global failure allowed
if (model.getSolver().isLearnOff() && F < 0) {
fails();
}
if (maxI > F) {
maxI = 0;
int lb, ub, i = 0;
// positive coefficients first
while (i < pos) {
if (I[i] - F > 0) {
lb = vars[i].getLB();
ub = lb + I[i];
if (vars[i].updateUpperBound(F + lb, this)) {
int nub = vars[i].getUB();
E += nub - ub;
I[i] = nub - lb;
}
}
if(maxI < I[i])maxI = I[i];
i++;
}
// then negative ones
while (i < l) {
if (I[i] - F > 0) {
lb = -vars[i].getUB();
ub = lb + I[i];
if (vars[i].updateLowerBound(-F - lb, this)) {
int nub = -vars[i].getLB();
E += nub - ub;
I[i] = nub - lb;
}
}
if(maxI < I[i])maxI = I[i];
i++;
}
}
if (E <= 0) {
this.setPassive();
}
}
/**
* Apply filtering when operator is GE
* @throws ContradictionException if contradiction is detected
*/
protected void filterOnGeq() throws ContradictionException {
int F = b - sumLB;
int E = sumUB - b;
// When explanations are on, no global failure allowed
if (model.getSolver().isLearnOff() && E < 0) {
fails();
}
if(maxI > E) {
maxI = 0;
int lb, ub, i = 0;
// positive coefficients first
while (i < pos) {
if (I[i] - E > 0) {
ub = vars[i].getUB();
lb = ub - I[i];
if (vars[i].updateLowerBound(ub - E, this)) {
int nlb = vars[i].getLB();
F -= nlb - lb;
I[i] = ub - nlb;
}
}
if(maxI < I[i])maxI = I[i];
i++;
}
// then negative ones
while (i < l) {
if (I[i] - E > 0) {
ub = -vars[i].getLB();
lb = ub - I[i];
if (vars[i].updateUpperBound(-ub + E, this)) {
int nlb = -vars[i].getUB();
F -= nlb - lb;
I[i] = ub - nlb;
}
}
if(maxI < I[i])maxI = I[i];
i++;
}
}
if (F <= 0) {
this.setPassive();
}
}
/**
* Apply filtering when operator is NE
* @throws ContradictionException if contradiction is detected
*/
protected void filterOnNeq() throws ContradictionException {
int F = b - sumLB;
int E = sumUB - b;
if (F < 0 || E < 0) {
setPassive();
return;
}
int w = -1;
int sum = 0;
for (int i = 0; i < l; i++) {
if (vars[i].isInstantiated()) {
sum += i < pos ? vars[i].getValue() : -vars[i].getValue();
} else if (w == -1) {
w = i;
} else return;
}
if (w == -1) {
if (sum == b) {
this.fails();
}
} else {
vars[w].removeValue(w < pos ? b - sum : sum - b, this);
}
}
@Override
public ESat isEntailed() {
int sumUB = 0, sumLB = 0, i = 0;
for (; i < pos; i++) { // first the positive coefficients
sumLB += vars[i].getLB();
sumUB += vars[i].getUB();
}
for (; i < l; i++) { // then the negative ones
sumLB -= vars[i].getUB();
sumUB -= vars[i].getLB();
}
return check(sumLB, sumUB);
}
/**
* Whether the current state of the scalar product is entailed
* @param sumLB sum of lower bounds
* @param sumUB sum of upper bounds
* @return the entailment check
*/
public ESat check(int sumLB, int sumUB){
switch (o) {
case NQ:
if (sumUB < b || sumLB > b) {
return ESat.TRUE;
}
if (sumUB == b && sumLB == b) {
return ESat.FALSE;
}
return ESat.UNDEFINED;
case LE:
if (sumUB <= b) {
return ESat.TRUE;
}
if (sumLB > b) {
return ESat.FALSE;
}
return ESat.UNDEFINED;
case GE:
if (sumLB >= b) {
return ESat.TRUE;
}
if (sumUB < b) {
return ESat.FALSE;
}
return ESat.UNDEFINED;
default:
if (sumLB == b && sumUB == b) {
return ESat.TRUE;
}
if (sumUB < b || sumLB > b) {
return ESat.FALSE;
}
return ESat.UNDEFINED;
}
}
@Override
public void explain(ExplanationForSignedClause explanation, ValueSortedMap front, Implications ig, int p) {
if (o == Operator.NQ) {
Propagator.defaultExplain(this, explanation, front, ig, p);
} else {
doExplain(explanation, front, ig, p);
}
}
void doExplain(ExplanationForSignedClause explanation, ValueSortedMap front, Implications ig, int p){
IntIterableRangeSet dom_before;
IntVar pivot = ig.getIntVarAt(p);
// first, compute F and E
int sumLB = 0;
int sumUB = 0;
int i = 0, lb, ub, la = 0, ua = 0, a = 0, ca = 0;
for (; i < pos; i++) { // first the positive coefficients
int f = front.getValue(vars[i]);
dom_before = ig.getDomainAt(f);
lb = dom_before.min();
ub = dom_before.max();
if (vars[i] == pivot) {
la = dom_before.min();
ua = dom_before.max();
a = i;
ca = 1;
}
sumLB += lb;
sumUB += ub;
}
for (; i < l; i++) { // then the negative ones
dom_before = ig.getDomainAt(front.getValue(vars[i]));
lb = -dom_before.max();
ub = -dom_before.min();
if (vars[i] == pivot) {
la = dom_before.min();
ua = dom_before.max();
a = i;
ca = -1;
}
sumLB += lb;
sumUB += ub;
}
int F = b - sumLB;
int E = sumUB - b;
if(ig.getDomainAt(p).isEmpty()){
doExplainGlobalFailure(explanation, front, ig, F, E);
return;
}
IntIterableRangeSet domain;
int la2 = IntIterableRangeSet.MIN, ua2 = IntIterableRangeSet.MAX;
if (a < pos) {
if (!o.equals(GE)) { // ie, LE or EQ
ua2 = F + la;
}
if (!o.equals(LE)) { // ie, GE or EQ
la2 = ua - E;
}
} else {
if (!o.equals(GE)) { // ie, LE or EQ
la2 = -F + ua;
}
if (!o.equals(LE)) { // ie, GE or EQ
ua2 = la + E;
}
}
domain = explanation.getRootSet(vars[a]);
if(la2 > ua2){
domain.clear();
}else {
domain.retainBetween(la2, ua2);
}
explanation.addLiteral(vars[a], domain, true);
i = 0;
for (; i < pos; i++) {
int min = IntIterableRangeSet.MIN;
int max = IntIterableRangeSet.MAX;
if (vars[i] != pivot) {
dom_before = ig.getDomainAt(front.getValue(vars[i]));
if (!o.equals(GE)) { // ie, LE or EQ
max = F + dom_before.min() - ca * (ca > 0 ? (ua2 + 1 - la) : (la2 - 1 - ua));
}
if (!o.equals(LE)) { // ie, GE or EQ
min = -E + dom_before.max() - ca * (ca > 0 ? la2 - 1 - ua : ua2 + 1 - la);
}
domain = explanation.getComplementSet(vars[i]);
if(o.equals(EQ)) {
assert max+1 <= min-1 : "empty range";
domain.removeBetween(max + 1, min - 1);
}else {
domain.retainBetween(min, max);
}
explanation.addLiteral(vars[i], domain, false);
}
}
for (; i < l; i++) {
int min = IntIterableRangeSet.MIN;
int max = IntIterableRangeSet.MAX;
if (vars[i] != pivot) {
dom_before = ig.getDomainAt(front.getValue(vars[i]));
if (!o.equals(GE)) { // ie, LE or EQ
min = -(F - dom_before.max() - ca * (ca > 0 ? ua2 + 1 - la : la2 - 1 - ua));
}
if (!o.equals(LE)) { // ie, GE or EQ
max = -(-E - dom_before.min() - ca * (ca > 0 ? la2 - 1 - ua : ua2 + 1 - la));
}
domain = explanation.getComplementSet(vars[i]);
if(o.equals(EQ)) {
assert max+1 <= min-1 : "empty range";
domain.removeBetween(max + 1, min - 1);
}else {
domain.retainBetween(min, max);
}
explanation.addLiteral(vars[i], domain, false);
}
}
}
void doExplainGlobalFailure(ExplanationForSignedClause explanation, ValueSortedMap front, Implications ig,
int F, int E) {
assert (F < 0) ^ (E < 0);
IntIterableRangeSet dom_before, domain;
int i = 0;
for (; i < pos; i++) {
int min = IntIterableRangeSet.MIN;
int max = IntIterableRangeSet.MAX;
dom_before = ig.getDomainAt(front.getValue(vars[i]));
if (F < 0) {
max = dom_before.min() - 1;
}else /*E < 0*/{
min = dom_before.max() + 1;
}
domain = explanation.getComplementSet(vars[i]);
domain.retainBetween(min, max);
explanation.addLiteral(vars[i], domain, false);
}
for (; i < l; i++) {
int min = IntIterableRangeSet.MIN;
int max = IntIterableRangeSet.MAX;
dom_before = ig.getDomainAt(front.getValue(vars[i]));
if (F < 0) { // ie, LE or EQ
min = dom_before.max() + 1;
}else /*E < 0*/{ // ie, GE or EQ
max = dom_before.min() - 1;
}
domain = explanation.getComplementSet(vars[i]);
domain.retainBetween(min, max);
explanation.addLiteral(vars[i], domain, false);
}
if(model.getSettings().explainGlobalFailureInSum() && !this.isReified()){
explainGlobal(explanation, front, ig, F, E);
}
}
protected void explainGlobal(ExplanationForSignedClause explanation, ValueSortedMap front, Implications ig,
int F, int E) {
assert (F < 0)^(E < 0);
IntIterableRangeSet dom_before;
IntIterableRangeSet domain;
int i = 0;
ClauseBuilder ngb = model.getClauseBuilder();
for (; i < l; i++) {
int min = IntIterableRangeSet.MIN;
int max = IntIterableRangeSet.MAX;
dom_before = ig.getDomainAt(front.getValue(vars[i]));
if (F < 0) {
// BEWARE // second part of the equation differs from non-global-fail case
if(i < pos) {
max = F + dom_before.min();
}else{
min = -(F - dom_before.max());
}
}else /*if (E < 0)*/ {
// BEWARE // second part of the equation differs from non-global-fail case
if(i < pos) {
min = -E + dom_before.max();
}else{
max = -(-E - dom_before.min());
}
}
domain = explanation.getRootSet(vars[i]);
domain = domain.duplicate();
domain.retainBetween(min, max);
ngb.put(vars[i], domain);
int k = 0;
for (; k < l; k++) {
if (k != i) {
min = IntIterableRangeSet.MIN;
max = IntIterableRangeSet.MAX;
dom_before = ig.getDomainAt(front.getValue(vars[k]));
if (F < 0) {
if(k < pos) {
min = dom_before.min();
}else{
max = dom_before.max();
}
}else /*if (E < 0)*/ {
if(k < pos) {
max = dom_before.max();
}else{
min = dom_before.min();
}
}
domain = explanation.getRootSet(vars[k]);
domain = domain.duplicate();
domain.removeBetween(min, max);
ngb.put(vars[k], domain.duplicate());
}
}
ngb.buildNogood(model);
if(E == -1 || F == -1)return; // the same nogood will be learned all the time
}
}
@Override
public String toString() {
StringBuilder linComb = new StringBuilder(20);
linComb.append(pos == 0 ? "-" : "").append(vars[0].getName());
int i = 1;
for (; i < pos; i++) {
linComb.append(" + ").append(vars[i].getName());
}
for (; i < l; i++) {
linComb.append(" - ").append(vars[i].getName());
}
linComb.append(" ").append(o).append(" ");
linComb.append(b);
return linComb.toString();
}
public static int nb(Operator co){
switch (co){
case LE:
return 1;
case GE:
return -1;
default:
return 0;
}
}
public static Operator nop(Operator co){
switch (co){
case LE:
return Operator.GE;
case GE:
return Operator.LE;
default:
return Operator.getOpposite(co);
}
}
protected PropSum opposite(){
return new PropSum(vars, pos, nop(o), b + nb(o));
}
}