Gurobi Optimizer Training

N
November
b 1313, 2009
About the instructor

 Greg
g Glockner
◦ PhD, Industrial Engineering, Georgia Tech
◦ Trained customers worldwide in modeling and
optimization software
◦ Joined Gurobi in spring of 2009
 Agenda

 Quick tour of Gurobi Optimizer

Q p

 Hands-on with Gurobi Interactive Shell

 Modeling with Gurobi Object-Oriented

Interfaces

 Working with the C Interface

 A quick
i k peek
k under
d the
h h
hood
d

18-Jun-10 © 2009 Gurobi Optimization 3

 Gurobi Optimizer: A Quick
Tour

18-Jun-10 © 2009 Gurobi Optimization 4

What is Gurobi Optimizer?

 Gurobi Optimizer
p is a high-end
g system
y for
solving linear programs and mixed-integer
programs

 If this
hi is
i unfamiliar
f ili to you, it’s
i ’ going
i to be
b a
very long day!
Algorithms in Gurobi

 LP: p
primal and dual simplex
p
 MIP: parallel branch and bound with
heuristics and cutting planes
◦ Both LP & MIP include automatic presolve
 IIS for LP and MIP

 Next release: parallel LP barrier with

crossover
 Interfaces to Gurobi

 Matrix
◦ C
 Object-oriented
◦ C++, Java, Python
◦ .NET through Microsoft Solver Foundation
 Interactive shell and command
command-line
line tool
 Modeling systems and MATLAB

 Other languages are possible but not

y supported
officially pp

18-Jun-10 © 2009 Gurobi Optimization 7

 File format support

 Models
◦ .LP, .MPS, .RLP, .REW
 Solution information
◦ .ILP, .MST, .BAS, .SOL
 Parameters
◦ .PRM
PRM

 All support built-in

built in compression (gzip,
(gzip bzip2)

18-Jun-10 © 2009 Gurobi Optimization 8

 Gurobi design

 Designed
g for efficiency
y
◦ Very efficient simplex
◦ Very powerful MIP
◦ Make
M k ththe mostt off multi-processors
lti

 Just enough customization

◦ Parameters
◦ Callbacks
◦ Some esoteric features deliberately left out to keep
Gurobi efficient

18-Jun-10 © 2009 Gurobi Optimization 9

 Command-line interface

 Very
y simple
p way
y to solve a model from a file
◦ gurobi_cl [param=value]* filename

 Useful for quick tests of models & parameters

 Does not save solution values

18-Jun-10 © 2009 Gurobi Optimization 10

 Try it: Command-line interface

 Open
p a command p
prompt
p

 cd to examples\data
p subdirectory
y of
Gurobi

 Solve these models:

◦ afiro.mps
◦ stein9.mps
stein9 mps
◦ misc07.mps

18-Jun-10 © 2009 Gurobi Optimization 11

 Log file (LP)
Optimize a model with 27 Rows, 32 Columns and 83 NonZeros

Presolve removed 21 rows and 23 columns

Presolve time: 0.00 sec.
Presolved: 6 Rows, 9 Columns, 26 Nonzeros

Iteration Objective Primal Inf. Dual Inf. Time

0 -4.8565680e+02 1.179278e+02 0.000000e+00 0s
3 -4.6475314e+02 0.000000e+00 0.000000e+00 0s

Solved in 3 iterations and 0.00 seconds

Optimal objective -4.647531429e+02

18-Jun-10 © 2009 Gurobi Optimization 12

 Log file (MIP)
Nodes | Current Node | Objective Bounds | Work
Expl
l Unexpl
l | Obj Depth
h IntInf
f | Incumbent
b BestBd
d G
Gap | It/Node
/ d Time
i

0 0 4.0000 0 6 5.0000 4.0000 20.0% - 0s

0 0 4.0000 0 8 5.0000 4.0000 20.0% - 0s
0 0 4.0000 0 8 5.0000 4.0000 20.0% - 0s
0 0 4.0000 0 8 5.0000 4.0000 20.0% - 0s

Cutting planes:
Gomory: 1

Explored 13 nodes (49 simplex iterations) in 0.00 seconds

Thread count was 2 (of 16 available processors)

Optimal solution found (tolerance 1.00e-04)

Best objective 5.0000000000e+00, best bound 5.0000000000e+00, gap 0.0%

18-Jun-10 © 2009 Gurobi Optimization 13

 Try it: Parameters

 Solve the model:

◦ misc07.mps

 Try different values between 0.0 and 1.0 for

the heuristics parameter

 Example
◦ gurobi
bi cl
l heuristics=0.2
h i ti 0 2 misc07.mps
i 07

18-Jun-10 © 2009 Gurobi Optimization 14

 Gurobi Parameters

 Parameters control the behavior of Gurobi

algorithms
◦ Termination criteria
◦ T l
Tolerances
◦ Simplex method
◦ MIP
◦ MIP cuts
◦ General
 All
ll parameters h
have good
dddefault
f l values
l
 Common parameters to adjust are
highlighted

18-Jun-10 © 2009 Gurobi Optimization 15

 Parameters: Termination

 Cutoff: Stop
p if objective
j is worse
 IterationLimit: Limit on simplex iterations
 NodeLimit: Limit on MIP nodes
 SolutionLimit: Limit on MIP solutions
 TimeLimit: Time limit in seconds

18-Jun-10 © 2009 Gurobi Optimization 16

 Parameters: Tolerances

 FeasibilityTol:
y Primal feasibility
y
 IntFeasTol: Integer feasibility
 MarkowitzTol: Simplex pivoting
 MIPGap: Relative MIP gap
 OptimalityTol: Dual feasibility

18-Jun-10 © 2009 Gurobi Optimization 17

 Parameters: Simplex

 LPMethod: Whether to use p primal or dual

 NormAdjust: Different norm pricing
 ObjScale: Value to rescale the objective
 PerturbValue: Value for perturbation
 Quad: Use quad precision
 ScaleFlag: Allow model scaling
 SimplexPricing: Variable pricing strategy

18-Jun-10 © 2009 Gurobi Optimization 18

 Parameters: MIP

 Heuristics: How frequently

q y to run heuristics
 NodefileDir: Directory to use for node files
 NodefileStart: When to start using node files
◦ No need to run out of memory for MIP tree
 RootMethod: LP method for root relaxation
 SubMIPNodes:
b d # nodes
d to use ffor RINS
 VarBranch: Branch variable selection

18-Jun-10 © 2009 Gurobi Optimization 19

 Parameters: MIP cuts

 Cuts: Global cut g

generation control
◦ Specific cuts can be controlled individually
 CutAggPasses: Number of constraint
aggregation passes in cut generation
 GomoryPasses: Number of Gomory cut passes

18-Jun-10 © 2009 Gurobi Optimization 20

 Parameters: Other

 Aggregate:
gg g Control p
presolve aggregation
gg g
 DisplayInterval: Logging frequency
 IISMethod: Tradeoff speed/size
p / in IIS
 LogfileName: Filename for logfile
 OutputFlag:
p g Whether to output
p to screen
 PreCrush: Turn on to add cut callbacks
 Presolve: Whether to use presolve
 Threads: # of threads to use for parallel

18-Jun-10 © 2009 Gurobi Optimization 21

 Documentation overview

 Getting
g Started Manual
 Reference Manual
◦ APIs
◦ Parameters
◦ Attributes
 Example Tour
◦ 16 examples for all supported languages

 Available both online and on your computer

18-Jun-10 © 2009 Gurobi Optimization 22

 Using Gurobi with AMPL

 Configuration
g
◦ Put Gurobi link files into AMPL directory
◦ Add AMPL directory to PATH
◦ Linux:
Li Add AMPL di directory
t tto LD
LD_LIBRARY_PATH
LIBRARY PATH

 When running, set solver to Gurobi

◦ option solver gurobi_ampl;

 Setting Gurobi parameters

◦ option gurobi_options $gurobi_options 'mipgap 0.01';

18-Jun-10 © 2009 Gurobi Optimization 23

 AMPL/Gurobi details

 See AMPL/Gurobi
/ User Guide for:
◦ Installation / Configuration information
◦ Full list of parameters (options)
 Di ib
Distributed
d with
i h AMPL/G
AMPL/Gurobi
bi

 N extra
No t licenses
li needed
d d
◦ Your existing AMPL license plus a Gurobi license
will allow y
you to run Gurobi from AMPL

18-Jun-10 © 2009 Gurobi Optimization 24

 Using Gurobi with GAMS

 When running,
g, set solver to Gurobi
◦ By default
 Set Gurobi as default solver during GAMS installation
◦ Manually
M ll
 Option LP = Gurobi; { or MIP or RMIP }

 Setting Gurobi parameters

◦ Via command-line flags
◦ Via options file

18-Jun-10 © 2009 Gurobi Optimization 25

 GAMS/Gurobi details

 See GAMS/Gurobi
/ Guide for:
◦ Installation / Configuration information
◦ Full list of options (parameters)
 A il bl on www.gams.com
Available

 A extra
An t license
li iis required
i d
◦ Separate licenses for Gurobi and GAMS do not
enable y
you to use Gurobi from GAMS
◦ Contact GAMS to obtain this license

18-Jun-10 © 2009 Gurobi Optimization 26

 Hands-on with the Gurobi
Interactive Shell

18-Jun-10 © 2009 Gurobi Optimization 27

Try it: Open the environment

 Open
p a command p
prompt
p

 Type:
◦ gurobi (Windows)
◦ gurobi.sh (Linux, Mac)

 To quit:
◦ Press Ctrl
Ctrl-Z
Z (Windows)
◦ Press Ctrl-D (Linux, Mac)
◦ Enter quit()
 Try it: Solve a model from a file

 cd to examples\data
p \ subdirectory
y of
Gurobi
 Open the Gurobi interactive shell
 Read a model
◦ model = read("afiro.mps")
 S l the
Solve th model
d l
◦ model.optimize()
 Print the solution
◦ model.printAttr("X")

18-Jun-10 © 2009 Gurobi Optimization 29

 Interactive Shell structure

 Interactive shell is a p
programming
g g language!
g g
◦ read() is a global function
◦ Model is an object
◦ Model.optimize() and d Model.printAttr() are
methods (functions on the Model object)

 So we can choose any name for the model:

model = read("afiro.mps") m = read("afiro.mps")
model.optimize() m.optimize()
model.printAttr("X") m.printAttr("X")

18-Jun-10 © 2009 Gurobi Optimization 30

 Attributes

 Attributes describe elements of a model

 Attributes pertain to a specific element

◦ Variable
◦ Constraint
◦ the overall Model
◦ Etc.

 Note that Attributes describe the model;

Parameters control the algorithms

18-Jun-10 © 2009 Gurobi Optimization 31

 Working with attributes

 Some attributes are read-onlyy ((ex: ObjVal);

j );
others are settable (ex: UB)
 Methods to access attributes
◦ getAttr(), setAttr(): use name or wildcards
◦ Directly: x.UB
 Examples:
◦ print m.getAttr("ObjVal")
◦ x.UB = 1.0

 Don't panic! Just refer to the Attributes

reference manual
18-Jun-10 © 2009 Gurobi Optimization 32
 Try it: Use model attributes

 Solve misc07.mps
p

 Print the following attributes:

◦ Objective value (ObjVal)
◦ Solution count (SolCount)

 Hint: use print keyword

18-Jun-10 © 2009 Gurobi Optimization 33

 Primary Gurobi classes

 Model the model

 Var a variable

 Constr a constraint

18-Jun-10 © 2009 Gurobi Optimization 34

 Selected global functions

 setParam()
()
◦ Set parameters for new models in this session

 read()
◦ Create a Model object from a model file

18-Jun-10 © 2009 Gurobi Optimization 35

 Variable methods

 Var.getAttr()
g ()
◦ Get attribute values
 Var.setAttr()
◦ Set attribute values

 E
Example:
l
 x.setAttr("VarName", "TotalCost")
 print x.RC

18-Jun-10 © 2009 Gurobi Optimization 36

 Constraint methods

 Constr.getAttr()
g ()
◦ Get attribute values
 Constr.setAttr()
◦ Set attribute values

 E
Example:
l
 c.setAttr("Sense", "=")
 print c.Slack

18-Jun-10 © 2009 Gurobi Optimization 37

 Key model methods

 Solving
g
 Attribute & parameter management
 Summary information
 Copying & Writing
 Model introspection
 Model modification
 Callbacks

18-Jun-10 © 2009 Gurobi Optimization 38

 Solving methods

 Model.optimize()
p ()
◦ Solve a model
 Model.presolve()
◦ Run presolve
 Model.reset()
◦ Reset solution information so that model is solved
from scratch
 Model.computeIIS()
p
◦ Compute LP or MIP IIS to detect infeasibilities

18-Jun-10 © 2009 Gurobi Optimization 39

 Management methods

 Model.getAttr()
g ()
◦ Get attribute values
 Model.setAttr()
◦ Set attribute values
 Model.setParam()
◦ Set parameter values for the model
 Model.getParamInfo()
◦ Find current parameter values
 Model.read()
◦ Read information for a model
◦ Parameter file, MIP starting values, etc.

18-Jun-10 © 2009 Gurobi Optimization 40

 Summary information

 Model.printAttr()
p ()
◦ Print nonzero attribute values

 Model.printStats()
◦ Print number of constraints, variables, etc.

 Model.printQuality()
◦ Print info about the solution quality

18-Jun-10 © 2009 Gurobi Optimization 41

 Copying & Writing

 Model.copy()
py()
◦ Return a copy of the model, ex: n = m.copy()
 Model.relax()
◦ Return a copy of the model where integer variables
are relaxed
 Model fixed()
Model.fixed()
◦ Create a copy of the model where integer variables
are fixed to values in the current solution

18-Jun-10 © 2009 Gurobi Optimization 42

 Model introspection

 Model.getVars()
g ()
◦ Get model variables as a list of Var objects
 Model.getConstrs()
◦ Get model constraints as a list of Constr objects
 Model.getRow()
◦ Get list of variables for a given constraint
 Model.getCol()
◦ Get list of constraints for a given variable

18-Jun-10 © 2009 Gurobi Optimization 43

 Model modification

 Model.remove()
()
◦ Remove an element (variable, constraint)
 Model.addVar()
◦ Return a new variable for the model
 Model.addConstr()
◦ Return a new constraint for the model
 Model.addRange()
◦ Return a new range constraint for the model
 Model.update()
◦ Process model updates (more about this later)

18-Jun-10 © 2009 Gurobi Optimization 44

 Try it: Use introspection

 Solve misc07.mps p
 Print variable names and values
◦ Code hint:
for v in [get the variables]:
print [variable name],[variable value]
◦ Notes
 v contains each variable value
 Must indent body of the for-loop
 Enter
E t blank
bl k li
line tto end
d th
the ffor-loop
l
 Extra credit: Just print nonzero values
◦ Code hint: if [variable is nonzero]:

18-Jun-10 © 2009 Gurobi Optimization 45

 Congratulations!

 You jjust wrote y

your first Python
y p
programs
g
using Gurobi!
◦ The Gurobi Interactive Shell is really just the Python
shell
h ll with
ith extensions
t i for
f Gurobi
G bi

 More in the next section

section…

18-Jun-10 © 2009 Gurobi Optimization 46

 Modeling with Gurobi OO
Interfaces

18-Jun-10 © 2009 Gurobi Optimization 47

3 Object-Oriented interfaces

 C++
 Java
 Python

 Similar in design
 All are efficient
◦ Each directly works with Gurobi matrices

 We will start with Python

◦ It
It’s
s interpreted
 What is Python?

 General p
purpose
p p
programming
g g language
g g
◦ Dynamic typing
◦ Automatic memory management
◦ Obj t
Object-oriented
i t d
◦ Uses indentation for block delimiters

 Includes a large standard library

18-Jun-10 © 2009 Gurobi Optimization 49

 Review: Key model methods

 Solving
g
 Attribute & parameter management
 Summary information
 Copying & Writing
 Model introspection
 Model modification
 Callbacks

18-Jun-10 © 2009 Gurobi Optimization 50

 Lazy updates

 Lazy
y updates
p make Gurobi interfaces efficient
◦ Changes are made in a batch
 Remember to call Model.update() when you
need to reference
f an updated model element
◦ Ex: After you add a variable, call Model.update()
before using the variable in a constraint
◦ Model.update() is called automatically inside
Model.optimize()

 For best performance, only call

Model update() when necessary
Model.update()

18-Jun-10 © 2009 Gurobi Optimization 51

 Basic steps to build a model

 Create a Model object

j
 Create the variables
 Call Model.update() once
 Create the constraints
 Call Model.optimize()
◦ No need to call Model.update() after adding
constraints – Model.optimize() calls Model.update()
auto at ca y
automatically

18-Jun-10 © 2009 Gurobi Optimization 52

 Building linear expressions

 The LinExpr
p class defines linear expressions
p

 Simple expressions: use the constructor

◦ x + 2 y: LinExpr([1.0, 2.0], [x, y])

 Complex
l expressions: add
dd terms
◦ z[1] + z[2] + … + z[n]:
sum = LinExpr()
for zi in z:
sum.addTerms(1.0, zi)

18-Jun-10 © 2009 Gurobi Optimization 53

 Try it: Build a simple LP

maximize x+y
subject to x+2y≤4
x, y ≥ 0
 Steps:
◦ Create the Model object
◦ M k maximization:
Make i i i set ModelSense
M d lS attribute
ib to -1
1
◦ Create the variables
◦ Update the model
◦ Add the constraints
◦ Solve & print objective
◦ W i model
Write d l using
i M d l i ("
Model.write("test.lp")
l ")

18-Jun-10 © 2009 Gurobi Optimization 54

 Try it: Add a constraint

x-y=0

 Steps:
◦ Add the constraint
◦ Solve & print objective

18-Jun-10 © 2009 Gurobi Optimization 55

 Adding a variable

 Tricky
y to add it to existing
g constraints

 Two options:
◦ Change coefficients in existing constraints
◦ Create a Column() object with the list of constraints
and coefficients

18-Jun-10 © 2009 Gurobi Optimization 56

 Try it: Add a new variable

maximize x+y+2z
subject to x+2y+z≤4
x–y=0
x, y, z ≥ 0
 Steps:
◦ Use Model.getConstrs()[0] to get 1st constraint
◦ Create Column() object for z constraint coefficients
◦ Add z variable
◦ Solve and print objective

18-Jun-10 © 2009 Gurobi Optimization 57

 Working with multiple models

 JJust create a Model object

j for each model!

 Example
m = read("model1.mps")
n = read("model2.mps")

18-Jun-10 © 2009 Gurobi Optimization 58

 Try it: Working with 2 models

 Steps:
p
◦ Read misc07.mps from examples\data subdirectory
◦ Copy it to another Model object
◦ S t solution
Set l ti limit
li it to
t 2 iin th
the first
fi t model
d l andd solve
l
◦ Save its solution as a MIP start (.MST file)
◦ Read the MIP start in the 2nd model object
j
◦ Solve it to optimality

18-Jun-10 © 2009 Gurobi Optimization 59

 Programming in Python

 Python
y is not jjust an interactive shell
◦ Store programs in files
◦ Run via gurobi.sh [filename]

 More care should be taken for a real program

◦ Error checking
◦ Testing the result codes of a model
 Optimal, feasible, etc.
◦ See examples\python subdirectory

18-Jun-10 © 2009 Gurobi Optimization 60

 Try it: Run a Python program

 Go to examples\python
p \py subdirectory
y
 Run some of the examples:
◦ gurobi workforce3.py
◦ gurobi diet.py
 Take a look at the source code

18-Jun-10 © 2009 Gurobi Optimization 61

 What about Java and C++?

 We used Python
y for training
g
◦ Nothing additional to setup and configure
◦ It is interactive – no compiling
◦ Easy
E to
t write
it – structure
t t iis lless rigid
i id

 Gurobi interfaces are similar for C++,

C++ Java

18-Jun-10 © 2009 Gurobi Optimization 62

 Naming differences

 In Python
y
◦ Classes: Model, Var, Constr, …
◦ Methods: getAttr(), setAttr()
 I JJava, C
In C++
◦ Classes: GRBModel, GRBVar, GRBConstr, …
◦ Methods: get(),
get() set()

 See Reference Manual for details

18-Jun-10 © 2009 Gurobi Optimization 63

 Other general differences

 Environment
◦ In Java and C++, start with an instance of GRBEnv
 Global parameters associated with the environment
◦ Constructor for GRBModel requires a GRBEnv object
 Error handling
◦ Must catch GRBException
◦ Mandatory
M d t ffor C
C++, JJava; optional
ti l ffor P
Python
th
 Global parameters
◦ GRBModel gets its copy of parameter settings
◦ Once GRBModel is instantiated, parameter changes in
GRBEnv are not reflected in model

18-Jun-10 © 2009 Gurobi Optimization 64

 Other C++ considerations

 Operator
p overloading
g
◦ Makes model easier to read & write

 Memory management
◦ Lack of garbage collector makes C++ programming
more complicated

18-Jun-10 © 2009 Gurobi Optimization 65

 C++ operator overloading

 Python
y
m.addConstr(
LinExpr([1, 2], [x, y]), '<', 5)
 C
C++
m.addConstr(x + 2*y <= 5);

 Helpful for simple expressions

18-Jun-10 © 2009 Gurobi Optimization 66

 C++ memory management tips

 Be careful about lifecycle

y of C++ objects
j
◦ Destructor is called when they go out of scope

 Delete pointers to C++ objects when finished

◦ Otherwise you will have a memory leak

 Gurobi creates some objects on the heap

◦ Ex: GRBModel::addVars()

 See examples\c++ for best practices

18-Jun-10 © 2009 Gurobi Optimization 67

 Comparing the OO interfaces

 Python
y
obj = m.getAttr("ObjVal")
nsol = m.getAttr("SolCount")
 J
Java
double obj = m.get(GRB.DoubleAttr.ObjVal)
int nsol = m.get(GRB.IntAttr.SolCount)
m get(GRB IntAttr SolCount)
 C++
double obj
j = m.get(GRB
g DoubleAttr ObjVal)
j
int nsol = m.get(GRB_IntAttr_SolCount)

18-Jun-10 © 2009 Gurobi Optimization 68

 Gurobi OO modeling tips

 Prototype
yp with Interactive Shell
◦ Object model is nearly identical
◦ Interactive and easy to test things
◦ Even
E callbacks
llb k are supportedt d

 Be careful with C++ memory management

18-Jun-10 © 2009 Gurobi Optimization 69

 Working
g with the C Interface

18-Jun-10 © 2009 Gurobi Optimization 70

Matrix interface

 Gurobi C++,, JJava and Python

y interfaces
represent a model via objects

 Gurobi C interface represents a model via

matrix coefficients
 Sparse matrices

 In most models,, nearly

y all entries in the A
matrix are zero

 Modern solvers like Gurobi only represent the

non-zero elements in the A matrix

 This compression is very efficient for memory

and computation but there is a learning curve

18-Jun-10 © 2009 Gurobi Optimization 72

 Compressed Sparse Matrices

 List the nonzero coefficients

 List the positions of these coefficients

◦ Remember that counting starts from zero!

 Two similar
l fformats to represent rows
(constraints) and columns (variables)

18-Jun-10 © 2009 Gurobi Optimization 73

 Sparse row example

18-Jun-10 © 2009 Gurobi Optimization 74

 Sparse column example

18-Jun-10 © 2009 Gurobi Optimization 75

 C interface functions

 Environment creation & destruction

 Model creation & modification
 Solution
 Model queries
 Input/output
 Attribute & parameter management
 Logging & callbacks
 Error handling

18-Jun-10 © 2009 Gurobi Optimization 76

 An example

maximize x + y + 2 z
subject to x + 2 y + 3 z <= 4
x + y >= 1
x, y, z binary

 This is mip1 from the examples subdirectory

18-Jun-10 © 2009 Gurobi Optimization 77

 Initialization

/* Create environment */
error = GRBloadenv(&env, "mip1.log");
if (error || env == NULL) {
fprintf(stderr, "Error: could not create environment\n");
exit(1);
exit(1)
}

/ Create an empty
/* p y model */
/
error = GRBnewmodel(env, &model, "mip1", 0, NULL, NULL, NULL,
NULL, NULL);
if (error) goto QUIT;

18-Jun-10 © 2009 Gurobi Optimization 78

 Creating variables

/* Add variables */
obj[0] = -1; obj[1] = -1; obj[2] = -2;
vtype[0] = GRB_BINARY; vtype[1] = GRB_BINARY; vtype[2] =
GRB_BINARY;
error = GRBaddvars(model,
GRBaddvars(model 3,
3 00, NULL
NULL, NULL,
NULL NULL,
NULL obj,
obj NULL,
NULL
NULL, vtype, NULL);
if (error) goto QUIT;

/* Integrate new variables */

error = GRBupdatemodel(model);
if (error) goto QUIT;

18-Jun-10 © 2009 Gurobi Optimization 79

 Adding constraints

/* First constraint: x + 2 y + 3 z <= 4 */

ind[0] = 0; ind[1] = 1; ind[2] = 2;
val[0] = 1; val[1] = 2; val[2] = 3;
error = GRBaddconstr(model, 3, ind, val, GRB_LESS_EQUAL, 4.0,
"c0");
c0 );
if (error) goto QUIT;

/* Second constraint: x + y >= 1 */

ind[0] = 0; ind[1] = 1;
val[0] = 1; val[1] = 1;
error = GRBaddconstr(model, 2, ind, val, GRB_GREATER_EQUAL, 1.0,
"c1");
"c1")
if (error) goto QUIT;

18-Jun-10 © 2009 Gurobi Optimization 80

 Optimization

/* Optimize model */
error = GRBoptimize(model);
if (error) goto QUIT;

/* Capture solution information */

error = GRBgetintattr(model, GRB_INT_ATTR_STATUS, &optimstatus);
if (error) goto QUIT;

error = GRBgetdblattr(model, GRB_DBL_ATTR_OBJVAL, &objval);

if (error) goto QUIT;

error = GRBgetdblattrarray(model, GRB_DBL_ATTR_X, 0, 3, sol);

if (error) goto QUIT;

18-Jun-10 © 2009 Gurobi Optimization 81

 Error reporting & cleanup

/* Error reporting */
if (error) {
printf("ERROR: %s\n", GRBgeterrormsg(env));
exit(1);
}

/* Free model */
( );
GRBfreemodel(model);

/* Free environment */
GRBfreeenv(env);

18-Jun-10 © 2009 Gurobi Optimization 82

 Comparing the interfaces

 Python
obj = m.getAttr("ObjVal")
nsol = m.getAttr("SolCount")
 Java
double obj = m.get(GRB.DoubleAttr.ObjVal)
int nsol = m.get(GRB.IntAttr.SolCount)
 C++
double obj = m.get(GRB_DoubleAttr_ObjVal)
int nsol = m.get(GRB_IntAttr_SolCount)
 C
error = GRBgetdblattr(m, GRB_DBL_ATTR_OBJVAL, &objval);
error = GRBgetintattr(m, GRB_INT_ATTR_SOLCOUNT, &nsol);

18-Jun-10 © 2009 Gurobi Optimization 83

 Summary

 C interface is the matrix interface to Gurobi

◦ Direct interface to Gurobi data structures

 OO interfaces are modeling interfaces

◦ Generally easier to use

 All are very lightweight

◦ Very low overhead
◦ All support efficient “lazy updates”
◦ Use your programming needs to pick an interface

18-Jun-10 © 2009 Gurobi Optimization 84