c++ debug gurobi, an uncaught exception error frequently occurs during debugging
AnsweredHello,
When I debug for a c++ gurobi problem, the following error frequently occurs although the codes are right.
libc++abi: terminating due to uncaught exception of type GRBException
Signal: SIGABRT (signal SIGABRT)
I use mac clion for the coding. Sometimes the project will back to normal after a while or after resetting the cache and reloading the project.
I find that sometimes gurobi C++ can't get the values of vars and output the above error during debugging although the model can be solved sucessfully after a while.
Do you know why?
Best,
Zhen
0
-
-
Gurobi Staff
Hi Zhen,
You should catch the exception as it is done in the Gurobi Examples, see, e.g., mip1_c++.cpp.
I hope this already helps. If not please share the Gurobi log output.Best,
Marika0 -
-
Thanks very much for the reply. I still can't find the reason that why sometimes runing sucessfully, while sometimes showing GRBException error. I guess may be due to the memory or cache issues since sometimes the output informatino is below.
Exception: EXC_BAD_ACCESS (code=1, address=0x100000008)
But I can't figure out which codes causing this.I put my full codes here:
#include "gurobi_c++.h"
// #include <iomanip> // for precision
#include <numeric>
// #include <unordered_set>
#include <vector>
class SingleProduct {
private:
// problem settings
double iniI = 0;
double iniCash = 0;
std::vector<double> meanDemands = {15, 15, 15,
15}; // std::vector<double>(4, 15);
std::string distribution_name = "poisson";
size_t T = 3; // meanDemands.size();
std::vector<double> unitVariOderCosts = std::vector<double>(T, 1);
std::vector<double> prices = std::vector<double>(T, 10);
double unitSalvageValue = 0.5;
std::vector<double> overheadCosts = std::vector<double>(T, 50);
double r0 = 0; // interest rate
double r1 = 0.1;
double r2 = 2;
double overdraftLimit = 500;
// sddp settings
int sampleNum = 2; // 10;
int forwardNum = 8; // 20;
int iterNum = 30;
double thetaInitialValue = -500;
public:
void solve() const;
};
void SingleProduct::solve() const {
const std::vector<int> sampleNums(T, sampleNum);
std::vector<std::vector<double>> sampleDetails(T);
// for (int t = 0; t < T; t++) {
// sampleDetails[t].resize(sampleNums[t]);
// auto sampling = Sampling(distribution_name, meanDemands[t]);
// sampleDetails[t] = sampling.generateSamples(sampleNums[t]);
// }
sampleDetails = {{5, 15}, {5, 15}, {5, 15}};
// 创建 Gurobi 环境与模型
GRBEnv env;
env.set(GRB_IntParam_OutputFlag, 0);
std::vector<GRBModel> models(T + 1, GRBModel(env));
// decision variables
std::vector<GRBVar> q(T);
std::vector<GRBVar> q_pre(T);
std::vector<GRBVar> theta(T - 1);
std::vector<GRBVar> I(T);
std::vector<GRBVar> B(T);
std::vector<GRBVar> cash(T);
std::vector<GRBVar> W0(T);
std::vector<GRBVar> W1(T);
std::vector<GRBVar> W2(T);
// build initial models for each stage
for (int t = 0; t < T + 1; t++) {
if (t > 0) {
cash[t - 1] =
models[t].addVar(-GRB_INFINITY, GRB_INFINITY, 0, GRB_CONTINUOUS,
"cash_" + std::to_string(t));
I[t - 1] = models[t].addVar(0, GRB_INFINITY, 0, GRB_CONTINUOUS,
"I_" + std::to_string(t));
B[t - 1] = models[t].addVar(0, GRB_INFINITY, 0, GRB_CONTINUOUS,
"B_" + std::to_string(t));
models[t].addConstr(I[t - 1] - B[t - 1] == 0);
if (t < T) {
models[t].addConstr(cash[t - 1] + prices[t - 1] * B[t - 1] == 0);
q_pre[t - 1] = models[t].addVar(0, GRB_INFINITY, 0, GRB_CONTINUOUS,
"q_pre_" + std::to_string(t + 1));
models[t].addConstr(q_pre[t - 1] == 0);
}
}
if (t < T) {
q[t] = models[t].addVar(0, GRB_INFINITY, 0, GRB_CONTINUOUS,
"q_" + std::to_string(t + 1));
W0[t] = models[t].addVar(0, GRB_INFINITY, 0, GRB_CONTINUOUS,
"W0_" + std::to_string(t + 1));
W1[t] = models[t].addVar(0, GRB_INFINITY, 0, GRB_CONTINUOUS,
"W1_" + std::to_string(t + 1));
W2[t] = models[t].addVar(0, GRB_INFINITY, 0, GRB_CONTINUOUS,
"W2_" + std::to_string(t + 1));
theta[t] =
models[t].addVar(-GRB_INFINITY, GRB_INFINITY, 0, GRB_CONTINUOUS,
"theta_" + std::to_string(t + 2));
models[t].addConstr(W1[t] <= overdraftLimit);
if (t == 0)
models[t].addConstr(iniCash - unitVariOderCosts[t] * q[t] - W0[t] +
W1[t] + W2[t] ==
overheadCosts[t]);
else {
models[t].addConstr(cash[t - 1] - unitVariOderCosts[t] * q[t] - W0[t] +
W1[t] + W2[t] ==
overheadCosts[t]);
}
models[t].addConstr(theta[t] >=
thetaInitialValue * (static_cast<double>(T) - t));
}
if (t == 0)
models[t].setObjective(overheadCosts[0] + unitVariOderCosts[0] * q[0] +
r2 * W2[0] + r1 * W1[0] - r0 * W0[0] + theta[0]);
models[t].update();
}
double intercepts[iterNum][T][forwardNum];
double slopes3[iterNum][T][forwardNum];
double slopes2[iterNum][T][forwardNum];
double slopes1[iterNum][T][forwardNum];
double qpreValues[iterNum][T][forwardNum];
double qValues[iterNum][T][forwardNum];
double IForwardValues[iterNum][T][forwardNum];
// double BForwardValues[iterNum][T][forwardNum];
// double cashForwardValues[iterNum][T][forwardNum];
double W0ForwardValues[iterNum][T][forwardNum];
double W1ForwardValues[iterNum][T][forwardNum];
double W2ForwardValues[iterNum][T][forwardNum];
int iter = 0;
while (iter < iterNum) {
// auto scenarioPaths =
// Sampling::generateScenarioPaths(forwardNum, sampleNums);
int scenarioPaths[8][3] = {{0, 0, 0}, {0, 0, 1}, {0, 1, 0}, {0, 1, 1},
{1, 0, 0}, {1, 0, 1}, {1, 1, 0}, {1, 1, 1}};
if (iter > 0) {
models[0].addConstr(
theta[0] >=
slopes1[iter - 1][0][0] * iniI +
slopes2[iter - 1][0][0] *
((1 + r0) * W0[0] - (1 + r1) * W1[0] - (1 + r2) * W2[0]) +
slopes3[iter - 1][0][0] * q[0] + intercepts[iter - 1][0][0]);
models[0].update();
}
try {
models[0].optimize();
} catch (const std::exception &e) {
std::cout << e.what() << std::endl;
}
// int optimstatus = models[0].get(GRB_IntAttr_Status);
// std::cout << "optimization status: " << optimstatus << std::endl;
// models[0].write("iter_" + std::to_string(iter + 1) + ".lp");
// models[0].write("iter_" + std::to_string(iter + 1) + ".sol");
for (int n = 0; n < forwardNum; n++) {
qValues[iter][0][n] = q[0].get(GRB_DoubleAttr_X);
W0ForwardValues[iter][0][n] = W0[0].get(GRB_DoubleAttr_X);
W1ForwardValues[iter][0][n] = W1[0].get(GRB_DoubleAttr_X);
W2ForwardValues[iter][0][n] = W2[0].get(GRB_DoubleAttr_X);
}
// forward
for (int t = 1; t < T + 1; t++) {
if (iter > 0 && t < T) {
std::vector<std::vector<double>> cutCoefficients(
forwardNum, std::vector<double>(4, 0));
for (int n = 0; n < forwardNum; n++) {
cutCoefficients[n][0] = slopes1[iter - 1][t][n];
cutCoefficients[n][1] = slopes2[iter - 1][t][n];
cutCoefficients[n][2] = slopes3[iter - 1][t][n];
cutCoefficients[n][3] = intercepts[iter - 1][t][n];
};
for (auto &finalCoefficient : cutCoefficients) { //
models[t].addConstr(
theta[t] >=
finalCoefficient[0] * (I[t - 1] + q_pre[t - 1]) +
finalCoefficient[1] *
((1 + r0) * W0[t] - (1 + r1) * W1[t] - (1 + r2) * W2[t]) +
finalCoefficient[2] * q[t] + finalCoefficient[3]);
}
}
for (int n = 0; n < forwardNum; n++) {
int index = scenarioPaths[n][t - 1];
double demand = sampleDetails[t - 1][index];
double rhs1 = t == 1 ? iniI - demand
: IForwardValues[iter][t - 1][n] +
qpreValues[iter][t - 2][n] - demand;
if (t < T) {
double rhs2 = prices[t - 1] * demand +
(1 + r0) * W0ForwardValues[iter][t - 1][n] -
(1 + r1) * W1ForwardValues[iter][t - 1][n] -
(1 + r2) * W2ForwardValues[iter][t - 1][n];
double rhs3 = qValues[iter][t - 1][n];
models[t].setObjective(
overheadCosts[t] + unitVariOderCosts[t] * q[t] -
prices[t - 1] * (demand - B[t - 1]) + r2 * W2[t] + r1 * W1[t] -
r0 * W0[t] + theta[t]);
models[t].getConstr(1).set(GRB_DoubleAttr_RHS, rhs2);
models[t].getConstr(2).set(GRB_DoubleAttr_RHS, rhs3);
} else
models[t].setObjective(-prices[t - 1] * (demand - B[t - 1]) -
unitSalvageValue * I[t - 1]);
models[t].getConstr(0).set(GRB_DoubleAttr_RHS, rhs1);
// optimize
try {
models[t].optimize();
} catch (const std::exception &e) {
std::cout << e.what() << std::endl;
}
IForwardValues[iter][t - 1][n] = I[t - 1].get(GRB_DoubleAttr_X);
// BForwardValues[t - 1][n] = B[t - 1].get(GRB_DoubleAttr_X);
// cashForwardValues[t - 1][n] = cash[t - 1].get(GRB_DoubleAttr_X);
if (t < T) {
qValues[iter][t][n] = q[t].get(GRB_DoubleAttr_X);
qpreValues[iter][t - 1][n] = q_pre[t - 1].get(GRB_DoubleAttr_X);
W0ForwardValues[iter][t][n] = W0[t].get(GRB_DoubleAttr_X);
W1ForwardValues[iter][t][n] = W1[t].get(GRB_DoubleAttr_X);
W2ForwardValues[iter][t][n] = W2[t].get(GRB_DoubleAttr_X);
}
}
}
// backward
std::vector interceptValues(T,
std::vector<std::vector<double>>(forwardNum));
std::vector slope1Values(T, std::vector<std::vector<double>>(forwardNum));
std::vector slope2Values(T, std::vector<std::vector<double>>(forwardNum));
std::vector slope3Values(T, std::vector<std::vector<double>>(forwardNum));
for (int t = 0; t < T; t++) {
for (int n = 0; n < forwardNum; n++) {
interceptValues[t][n].resize(sampleNums[t]);
slope1Values[t][n].resize(sampleNums[t]);
slope2Values[t][n].resize(sampleNums[t]);
slope3Values[t][n].resize(sampleNums[t]);
}
}
for (size_t t = T; t > 0; t--) {
for (int n = 0; n < forwardNum; n++) {
size_t S = sampleDetails[t - 1].size();
for (size_t s = 0; s < S; s++) {
auto demand = sampleDetails[t - 1][s];
double rhs1 = t == 1 ? iniI - demand
: IForwardValues[iter][t - 1][n] +
qpreValues[iter][t - 2][n] - demand;
if (t < T) {
double rhs2 = prices[t - 1] * demand +
(1 + r0) * W0ForwardValues[iter][t - 1][n] -
(1 + r1) * W1ForwardValues[iter][t - 1][n] -
(1 + r2) * W2ForwardValues[iter][t - 1][n];
double rhs3 = qValues[iter][t - 1][n];
models[t].setObjective(
overheadCosts[t] + unitVariOderCosts[t] * q[t] -
prices[t - 1] * (demand - B[t - 1]) + r2 * W2[t] + r1 * W1[t] -
r0 * W0[t] + theta[t]);
models[t].getConstr(1).set(GRB_DoubleAttr_RHS, rhs2);
models[t].getConstr(2).set(GRB_DoubleAttr_RHS, rhs3);
} else
models[t].setObjective(-prices[t - 1] * (demand - B[t - 1]) -
unitSalvageValue * I[t - 1]);
models[t].getConstr(0).set(GRB_DoubleAttr_RHS, rhs1);
// optimize
try {
models[t].optimize();
} catch (const std::exception &e) {
std::cout << e.what() << std::endl;
}
// if (iter == 3 and t == 1) {
// models[t].write("iter_" + std::to_string(iter + 1) + "_sub_" +
// std::to_string(t) + "^" + std::to_string(n + 1) +
// ".lp");
// void();
// }
int piNum = models[t].get(GRB_IntAttr_NumConstrs);
double pi[piNum];
double rhs[piNum];
for (int p = 0; p < piNum; p++) {
GRBConstr constraint = models[t].getConstr(p);
pi[p] = constraint.get(GRB_DoubleAttr_Pi);
rhs[p] = constraint.get(GRB_DoubleAttr_RHS);
}
if (t < T) {
interceptValues[t - 1][n][s] +=
-pi[0] * demand + pi[1] * prices[t - 1] * demand -
prices[t - 1] * demand + overheadCosts[t];
} else {
interceptValues[t - 1][n][s] +=
-pi[0] * demand - prices[t - 1] * demand;
}
for (size_t k = 3; k < piNum; k++)
interceptValues[t - 1][n][s] += pi[k] * rhs[k];
slope1Values[t - 1][n][s] = pi[0];
if (t < T) {
slope2Values[t - 1][n][s] = pi[1];
slope3Values[t - 1][n][s] = pi[2];
}
}
double avgIntercept;
double avgSlope1;
double avgSlope2;
double avgSlope3;
for (size_t s = 0; s < S; s++) {
double sum = std::accumulate(interceptValues[t - 1][n].begin(),
interceptValues[t - 1][n].end(), 0.0);
avgIntercept = sum / static_cast<double>(S);
sum = std::accumulate(slope1Values[t - 1][n].begin(),
slope1Values[t - 1][n].end(), 0.0);
avgSlope1 = sum / static_cast<double>(S);
sum = std::accumulate(slope2Values[t - 1][n].begin(),
slope2Values[t - 1][n].end(), 0.0);
avgSlope2 = sum / static_cast<double>(S);
sum = std::accumulate(slope3Values[t - 1][n].begin(),
slope3Values[t - 1][n].end(), 0.0);
avgSlope3 = sum / static_cast<double>(S);
}
slopes1[iter][t - 1][n] = avgSlope1;
slopes2[iter][t - 1][n] = avgSlope2;
slopes3[iter][t - 1][n] = avgSlope3;
intercepts[iter][t - 1][n] = avgIntercept;
}
}
// std::cout << "iteration " << iter << ", objective is " << std::fixed
// << std::setprecision(2) <<
// -models[0].get(GRB_DoubleAttr_ObjVal)
// << std::endl;
iter = iter + 1;
}
std::cout << "********************************************" << std::endl;
double finalValue = -models[0].get(GRB_DoubleAttr_ObjVal);
double Q1 = qValues[iter - 1][0][0];
std::cout << "after " << iter << " iterations: " << std::endl;
std::cout << "final expected cash balance is " << finalValue << std::endl;
std::cout << "ordering Q in the first period is " << Q1 << std::endl;
}
int main() {
const SingleProduct singleProduct;
const auto start_time = std::chrono::high_resolution_clock::now();
singleProduct.solve();
const auto end_time = std::chrono::high_resolution_clock::now();
const std::chrono::duration<double> diff = end_time - start_time;
std::cout << "cpu time is: " << diff.count() << " seconds" << std::endl;
return 0;
}0 -
-
-
It is not enough to only catch the optimize call.
When testing your code, it happens that solution attributes are accessed for an infeasible or unbounded model. Before accessing the X attribute, you should make sure that a solution exists, for example, by checking the attribute SolCount.
Another issue I observed was with linemodels[t].getConstr(0).set(GRB_DoubleAttr_RHS, rhs1);
Here, the rhs1 was -nan in one case, which also throws an error.So, you need to do a much more detailed exception handling in your code!0 -
-
Thanks Marika very much for the reply.
The weired thing is that the codes sometimes run sucessfully while sometimes not. If there is something wrong in the model setting, it should fail every time.
I don't know whether the codes failed every time when you run my code or can sometimes run successfully.
Best,
Zhen
0 -
-
For example, one model is running sucessfully, the following code get value 1,
models[0].get(GRB_IntAttr_SolCount)
but very strange that it can't get the value of its vars
0 -
-
-
Why do you think the problem in this specific case occurs when accessing the variable values?
As I said, there are several issues with your code.
You must handle exceptions properly! Then, you can see which method failed.0 -
-
Thanks very much Marika! Your comments gave me the inspiration to check my codes.
After checking for several hours, I found the issue: the definition of a decision variable
std::vector<GRBVar> theta(T);
The number of decision variable 'theta' should be T not T-1 in the previous code. After revising, every time it runs smoothly now.
Debugging in C++ is much more complex than Python as the error notice is not that detailed and close to the real error codes.
0 -
Please sign in to leave a comment.
Comments
7 comments