MIPNLP stoaMINLP - Revision history

Bjorn: /* Prob structure */

2015-03-09T14:34:29Z

Prob structure

← Older revision		Revision as of 09:34, 9 March 2015
Line 100:		Line 100:
	\|''x_0''\|\|Starting point.		\|''x_0''\|\|Starting point.
	\|-valign="top"		\|-valign="top"
	\|''Convex''\|\|If Convex==1, assume NLP problems are convex, and only one local NLP solver call is used at each node. <br>If Convex==0 (Default), multiMin is used to do many calls to a local solver to determine the global minima at each node. The global minimum with most components integer valued is chosen.		\|''Convex''\|\|If Convex==1, assume NLP problems are convex, and only one local NLP solver call is used at each node. <br>If Convex==0 (Default), [[multiMin]] is used to do many calls to a local solver to determine the global minima at each node. The global minimum with most components integer valued is chosen.
	\|-valign="top"		\|-valign="top"
	\|''MaxCPU''\|\|Maximal CPU Time (in seconds) to be used by stoaMINLP, stops with best point found		\|''MaxCPU''\|\|Maximal CPU Time (in seconds) to be used by stoaMINLP, stops with best point found
Line 106:		Line 106:
	\|''PriLev''\|\|Print level in stoaMINLP (default 1). Also see optParam.IterPrint		\|''PriLev''\|\|Print level in stoaMINLP (default 1). Also see optParam.IterPrint
	\|-valign="top"		\|-valign="top"
	\|''PriLevOpt''\|\|Print level in sub-solvers (SNOPT and other NLP solvers):<br>=0 No output; ''>''0 Convergence results.<br>''>''1 Output every iteration, <br>''>''2 Output each step in the NLP alg<br>For other NLP solvers, see the documentation for the solver		\|''PriLevOpt''\|\|Print level in sub-solvers ([[SOL_SNOPT\|SNOPT]] and other NLP solvers):<br>=0 No output; ''>''0 Convergence results.<br>''>''1 Output every iteration, <br>''>''2 Output each step in the NLP alg<br>For other NLP solvers, see the documentation for the solver
	\|-valign="top"		\|-valign="top"
	\|''WarmStart''\|\|If true, ''>''0, stoaMINLP reads the output from the last run from Prob.stoaMINLP, if it exists. If it doesn't exist, stoaMINLP attempts to open and read warm start data from mat-file stoaMINLPSave.mat. stoaMINLP uses the warm start information to continue from the last run. The mat-file minlp- SolveSave.mat is saved every Prob.MIP.SaveFreq iteration.		\|''WarmStart''\|\|If true, ''>''0, stoaMINLP reads the output from the last run from Prob.stoaMINLP, if it exists. If it doesn't exist, stoaMINLP attempts to open and read warm start data from mat-file stoaMINLPSave.mat. stoaMINLP uses the warm start information to continue from the last run. The mat-file minlp- SolveSave.mat is saved every Prob.MIP.SaveFreq iteration.
	\|-valign="top"		\|-valign="top"
	\|''SolverNLP''\|\|Name of the solver used for NLP subproblems. If empty, the default solver is found calling GetSolver('con',1); <br>If TOMLAB /SOL installed, SNOPT is the default solver.<br>If SolverNLP is a SOL solver (SNOPT, MINOS or NPSOL), the SOL.optPar and SOL.PrintFile is used:<br>See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters		\|''SolverNLP''\|\|Name of the solver used for NLP subproblems. If empty, the default solver is found calling GetSolver('con',1); <br>If TOMLAB /SOL installed, [[SOL_SNOPT\|SNOPT]] is the default solver.<br>If SolverNLP is a SOL solver ([[SOL_SNOPT\|SNOPT]], [[SOL_MINOS\|MINOS]] or [[SOL_NPSOL\|NPSOL]]), the SOL.optPar and SOL.PrintFile is used:<br>See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters
	\|-		\|-
	\|''SolverNLP0''\|\|Name of the solver used for the initial NLP-problem. If empty, SolverNLP is used.		\|''SolverNLP0''\|\|Name of the solver used for the initial NLP-problem. If empty, SolverNLP is used.
	\|-valign="top"		\|-valign="top"
	\|''SolverLP''\|\|Name of the solver used for LP subproblems. If empty, the default solver is found calling GetSolver('lp',1). If TOMLAB /SOL or /SNOPT or /NPSOL installed, MINOS is the default solver. If SolverLP is a SOL solver (MINOS, SNOPT or NPSOL), the SOL.optPar and SOL.PrintFile is used: See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters.		\|''SolverLP''\|\|Name of the solver used for LP subproblems. If empty, the default solver is found calling GetSolver('lp',1). If TOMLAB /[[SOL]] or /[[SOL_SNOPT\|SNOPT]] or /[[SOL_NPSOL\|NPSOL]] installed, [[SOL_MINOS\|MINOS]] is the default solver. If SolverLP is a [[SOL]] solver ([[SOL_MINOS\|MINOS]], [[SOL_SNOPT\|SNOPT]] or [[SOL_NPSOL\|NPSOL]]), the SOL.optPar and SOL.PrintFile is used: See the wiki-manuals or the in-MATLAB help of minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters.
	\|-valign="top"		\|-valign="top"
	\|''RandState''\|\|See help rngset for how to initialize the random generator. <br>Default is RandState = -1<br>If Convex == 0 and globalSolver == 'multiMin', RandState is sent to multiMin to initizalize the random generator.		\|''RandState''\|\|See help rngset for how to initialize the random generator. <br>Default is RandState = -1<br>If Convex == 0 and globalSolver == 'multiMin', RandState is sent to [[multiMin]] to initizalize the random generator.
	\|-valign="top"		\|-valign="top"
	\|''xInit''\|\|Parameter sent to the solver multiMin which is used to solve relaxed sub-problems for nonconvex problems (with Prob.Convex set to false). xInit determines the way initial points are generated and is set as follows:<br>		\|''xInit''\|\|Parameter sent to the solver [[multiMin]] which is used to solve relaxed sub-problems for nonconvex problems (with Prob.Convex set to false). xInit determines the way initial points are generated and is set as follows:<br>
	<br>		<br>
	Either<br>		Either<br>
Line 129:		Line 129:
	Default value for xInit is min(3000,max(60,N*6)) at all nodes except the root node<br>		Default value for xInit is min(3000,max(60,N*6)) at all nodes except the root node<br>
	\|-valign="top"		\|-valign="top"
	\|''xInit0''\|\|Parameter sent to the solver multiMin when solving the initial relaxed NLP problem at the root node.<br>		\|''xInit0''\|\|Parameter sent to the solver [[multiMin]] when solving the initial relaxed NLP problem at the root node.<br>
	Default value for xInit0 is min(3000,max(100,N*10))<br>		Default value for xInit0 is min(3000,max(100,N*10))<br>
	See xInit for more information.<br>		See xInit for more information.<br>

Bjorn: /* optParam structure */

2015-02-13T12:30:33Z

optParam structure

← Older revision		Revision as of 07:30, 13 February 2015
Line 238:		Line 238:
	The information printed for different levels is as follows: <br>		The information printed for different levels is as follows: <br>

	>0 ~~A line~~ at each ~~evaluated~~ node with ~~the following~~: <br>		>0 Two lines at each node with an improved integer solution: <br>
	Iteration number: <br>		Iteration number: <br>
	Depth in tree (symbol L[] - empty list, symbol Gap - Dual Gap convergence), <br>fNLP (Optimal f(x) current node), <br>fIPMin (Best integer feasible f(x) found), <br>LowBnd (Lower bound on optimal integer feasible f(x)), <br>Dual Gap in absolut value and percent, <br>The length of the node list L, -L-, <br>The Inform and ExitFlag the solver returned at the current node, <br>FuEv (Number of function evaluations used by solver at current node), <br>~~date/~~time stamp. <br>Outer approximation linearizations, number of cuts in active/inactive pool (\|C\| #A #I). <br>		* Depth in tree (symbol L[] - empty list, symbol Gap - Dual Gap convergence)<br>
			* fNLP (Optimal f(x) current node)<br>
			* fIPMin (Best integer feasible f(x) found)<br>
			* LowBnd (Lower bound on optimal integer feasible f(x))<br>
			* Dual Gap in absolut value and percent<br>
			* The length of the node list L, -L-<br>
			* The Inform and ExitFlag the solver returned at the current node<br>
			* FuEv (Number of function evaluations used by solver at current node)<br>
			* Date and time stamp.<br>
			* Outer approximation linearizations, number of cuts in active/inactive pool (\|C\| #A #I). <br>

	> 1 ~~A line~~ at each evaluated node with the information described above.<br>		> 1 Two lines at each evaluated node with the information described above.<br>
	Moreover, with IterPrint enabled a notice is printed whenever the algorithm switches node selection methods. <br>		Moreover, with IterPrint enabled a notice is printed whenever the algorithm switches node selection methods. <br>
	Setting a higher value of IterPrint gives more detailed information. <br>		Setting a higher value of IterPrint gives more detailed information. <br>
Line 250:		Line 259:
	>4 Sets POOLPRINT = 1, print info about cut-pool management (if used).		>4 Sets POOLPRINT = 1, print info about cut-pool management (if used).
	\|-valign="top"		\|-valign="top"
	\|''bTol''\|\|Linear constraint violation convergence tolerance.		\|''bTol''\|\|Linear constraint violation convergence tolerance. Default 1.0e-6
	\|-valign="top"		\|-valign="top"
	\|''cTol''\|\|Constraint violation convergence tolerance.		\|''cTol''\|\|Constraint violation convergence tolerance. Default 1.0e-6
	\|-valign="top"		\|-valign="top"
	\|''fTol''\|\|Optimality tolerance.		\|''fTol''\|\|Optimality tolerance. Default 3.0e-13
			\|-valign="top"
			\|''xTol''\|\|Variable convergence tolerance. Default 2.2204e-13
	\|}		\|}

Bjorn: /* STOAMINLP structure */

2015-02-13T12:24:17Z

STOAMINLP structure

← Older revision		Revision as of 07:24, 13 February 2015
Line 221:		Line 221:
	\|''PrintLin''\|\|Set nonzero to print short information about the type of linearization added. Default 0. Enabled when IterPrint > 3.		\|''PrintLin''\|\|Set nonzero to print short information about the type of linearization added. Default 0. Enabled when IterPrint > 3.
	\|-valign="top"		\|-valign="top"
	\|''PRINTPOOL''\|\|Set nonzero to print ~~information about~~ the cutpool management. Default 0. Enabled when IterPrint > 4.		\|''PRINTPOOL''\|\|Set nonzero to print Set nonzero to print the decisions taken in the cutpool management.Default 0. Enabled when IterPrint > 4.
	\|}		\|}

Bjorn: /* STOAMINLP structure */

2015-02-13T12:22:34Z

STOAMINLP structure

← Older revision		Revision as of 07:22, 13 February 2015
Line 187:		Line 187:
	\|'''Field'''\|\|'''Description'''		\|'''Field'''\|\|'''Description'''
	\|-valign="top"		\|-valign="top"
	\|''~~USECUTS~~''\|\|Flag to use cutting plane algorithm described in [[StoaMINLP#Algorithm]]. Using cuts is not recommended for nonconvex problems. <br>Set nonzero to enable, zero to disable. By default, ~~USECUTS =~~ Prob.Convex.		\|''USE_CUTS''\|\|Flag to use cutting plane algorithm described in [[StoaMINLP#Algorithm]]. Using cuts is not recommended for nonconvex problems. <br>Set nonzero to enable, zero to disable. By default, STOAMINLP.USE_CUTS is set to Prob.Convex.
	\|-valign="top"		\|-valign="top"
	\|''~~B_ecp~~''\|\|Parameter regulating likelihood of linearization using extended cutting plane. See [[StoaMINLP#Algorithm]] for more information. Default 10.		\|''LoL_ecp''\|\|Parameter regulating likelihood of linearization using extended cutting plane. See [[StoaMINLP#Algorithm]] for more information. Default 10.
	\|-valign="top"		\|-valign="top"
	\|''~~K_ecp~~''\|\|Parameter regulating "tailing off" detection for extended cutting plane linearizations. See [[StoaMINLP#Algorithm]] for more information. Default 0.001.		\|''TOD_ecp''\|\|Parameter regulating "tailing off" detection for extended cutting plane linearizations. See [[StoaMINLP#Algorithm]] for more information. Default 0.001.
	\|-valign="top"		\|-valign="top"
	\|''~~M_ecp~~''\|\|Parameter regulating the balance between branching the search tree and adding linearizations using extended cutting plane. See [[StoaMINLP#Algorithm]] for more information. Default 10.		\|''BCB_ecp''\|\|Parameter regulating the balance between branching the search tree and adding linearizations using extended cutting plane. See [[StoaMINLP#Algorithm]] for more information. Default 10.
	\|-valign="top"		\|-valign="top"
	\|''~~B_ff~~''\|\|Parameter regulating likelihood of linearization using fixed fractional integer decision variables. See [[StoaMINLP#Algorithm]] for more information. Default 10.		\|''LoL_ff''\|\|Parameter regulating likelihood of linearization using fixed fractional integer decision variables. See [[StoaMINLP#Algorithm]] for more information. Default 10.
	\|-valign="top"		\|-valign="top"
	\|''~~K_ff~~''\|\|Parameter regulating "tailing off" detection for linearizations using fixed fractional integer decision variables. See [[StoaMINLP#Algorithm]] for more information. Default 0.001.		\|''TOD_ff''\|\|Parameter regulating "tailing off" detection for linearizations using fixed fractional integer decision variables. See [[StoaMINLP#Algorithm]] for more information. Default 0.001.
	\|-valign="top"		\|-valign="top"
	\|''~~M_ff~~''\|\|Parameter regulating the balance between branching the search tree and adding linearizations using fixed fractional integer decision variables. See [[StoaMINLP#Algorithm]] for more information. Default 10.		\|''BCB_ff''\|\|Parameter regulating the balance between branching the search tree and adding linearizations using fixed fractional integer decision variables. See [[StoaMINLP#Algorithm]] for more information. Default 10.
	\|-valign="top"		\|-valign="top"
	\|''~~B_nlpr~~''\|\|Parameter regulating likelihood of linearization using the solution of the relaxed NLP problem. See [[StoaMINLP#Algorithm]] for more information. Default 10.		\|''LoL_nlpr''\|\|Parameter regulating likelihood of linearization using the solution of the relaxed NLP problem. See [[StoaMINLP#Algorithm]] for more information. Default 10.
	\|-valign="top"		\|-valign="top"
	\|''~~K_nlpr~~''\|\|Parameter regulating "tailing off" detection for linearizations using the solution of the relaxed NLP problem. See [[StoaMINLP#Algorithm]] for more information. Default 0.001.		\|''TOD_nlpr''\|\|Parameter regulating "tailing off" detection for linearizations using the solution of the relaxed NLP problem. See [[StoaMINLP#Algorithm]] for more information. Default 0.001.
	\|-valign="top"		\|-valign="top"
	\|''~~M_nlpr~~''\|\|Parameter regulating the balance between branching the search tree and adding linearizations using the solution of the relaxed NLP problem. See [[StoaMINLP#Algorithm]] for more information. Default 10.		\|''BCB_nlpr''\|\|Parameter regulating the balance between branching the search tree and adding linearizations using the solution of the relaxed NLP problem. See [[StoaMINLP#Algorithm]] for more information. Default 10.
	\|-valign="top"		\|-valign="top"
	\|''rTYPE''\|\|= 0 r is fixed to the value in rVALUE.<br>= 1 r is a uniformly distributed random number in the interval [0,1]. Default.		\|''rTYPE''\|\|= 0 r is fixed to the value in rVALUE.<br>= 1 r is a uniformly distributed random number in the interval [0,1]. Default.
Line 213:		Line 213:
	\|''CUTPOOL''\|\|= 0 All linearizations constraints are kept in the master problem. <br>= 1 Manage a pool with inactive linearization constraints. If an inactive constraint gets violated, it is brought back from the pool. Default.		\|''CUTPOOL''\|\|= 0 All linearizations constraints are kept in the master problem. <br>= 1 Manage a pool with inactive linearization constraints. If an inactive constraint gets violated, it is brought back from the pool. Default.
	\|-valign="top"		\|-valign="top"
	\|''~~CUTDELBND~~''\|\|The number of consecutive times a linearization can be inactive before it gets removed and put into the inactive pool. Default 15.		\|''CutDelBnd''\|\|The number of consecutive times a linearization can be inactive before it gets removed and put into the inactive pool. Default is 15 times.
	\|-valign="top"		\|-valign="top"
	\|''~~CUTLIMDEL~~''\|\|= 0 No extra deletion of cuts. Default when Prob.Convex == 1.<br>= 1 Delete any cuts limiting the true feasible region ~~when~~ detected. Such cuts will appear if the problem has nonlinear constraints. Default when Prob.Convex == 0.		\|''LIM_DEL_CUT''\|\|= 0 No extra deletion of cuts. Default when Prob.Convex == 1.<br>= 1 Delete any cuts limiting the true feasible region whenever detected. Such cuts will appear if the problem has nonlinear constraints. Default when Prob.Convex == 0.
	\|-valign="top"		\|-valign="top"
	\|''~~PCITL~~''\|\|Limit on time spent per node calculating initial pseudo costs when VarSel == 3. Default is 120 (s).		\|''PsCoItLim''\|\|Limit on time spent per node calculating initial pseudo costs when MIP.VarSel == 3. Default is 120 (s).
	\|-valign="top"		\|-valign="top"
	\|''~~LINPRINT~~''\|\|Set nonzero to print short information about the type of linearization added. Default 0. ~~enabled~~ when IterPrint > 3.		\|''PrintLin''\|\|Set nonzero to print short information about the type of linearization added. Default 0. Enabled when IterPrint > 3.
	\|-valign="top"		\|-valign="top"
	\|''~~POOLPRINT~~''\|\|Set nonzero to print information about the cutpool management. Default 0. Enabled when IterPrint > 4.		\|''PRINTPOOL''\|\|Set nonzero to print information about the cutpool management. Default 0. Enabled when IterPrint > 4.
	\|}		\|}

Bjorn: /* Prob structure */

2015-02-13T12:16:17Z

Prob structure

← Older revision		Revision as of 07:16, 13 February 2015
Line 116:		Line 116:
	\|''SolverLP''\|\|Name of the solver used for LP subproblems. If empty, the default solver is found calling GetSolver('lp',1). If TOMLAB /SOL or /SNOPT or /NPSOL installed, MINOS is the default solver. If SolverLP is a SOL solver (MINOS, SNOPT or NPSOL), the SOL.optPar and SOL.PrintFile is used: See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters.		\|''SolverLP''\|\|Name of the solver used for LP subproblems. If empty, the default solver is found calling GetSolver('lp',1). If TOMLAB /SOL or /SNOPT or /NPSOL installed, MINOS is the default solver. If SolverLP is a SOL solver (MINOS, SNOPT or NPSOL), the SOL.optPar and SOL.PrintFile is used: See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters.
	\|-valign="top"		\|-valign="top"
	\|''RandState''\|\|~~If Convex == 0, RandState is sent to multiMin~~ to initialize the random generator. ~~RandState is used as follows:~~ <br>~~If ''> ''0, rand('state',~~RandState~~) is set to initialize the pseudo~~-~~random generator.~~<br> If ~~''< ''~~0~~, rand(~~'~~state~~',~~sum(100*clock))~~ is ~~set~~ to ~~give a new set of~~ random ~~values each run. <br>If RandState == 0, rand('state',) is not called~~. ~~<br>Default is RandState = -1~~		\|''RandState''\|\|See help rngset for how to initialize the random generator. <br>Default is RandState = -1<br>If Convex == 0 and globalSolver == 'multiMin', RandState is sent to multiMin to initizalize the random generator.
	\|-valign="top"		\|-valign="top"
	\|''xInit''\|\|Parameter sent to the solver multiMin which is used to solve relaxed sub-problems for nonconvex problems (with Prob.Convex set to false). xInit determines the way initial points are generated and is set as follows:<br>		\|''xInit''\|\|Parameter sent to the solver multiMin which is used to solve relaxed sub-problems for nonconvex problems (with Prob.Convex set to false). xInit determines the way initial points are generated and is set as follows:<br>

Bjorn: /* MIP structure */

2015-02-11T15:38:53Z

MIP structure

← Older revision		Revision as of 10:38, 11 February 2015
Line 151:		Line 151:
	\|''IntVars''\|\|If empty, all variables are assumed non-integer. <br>If islogical(IntVars) (=all elements are 0/1), then 1 = integer variable, 0 = continuous variable. <br>If any element ''>''1, IntVars is the indices for integer variables.		\|''IntVars''\|\|If empty, all variables are assumed non-integer. <br>If islogical(IntVars) (=all elements are 0/1), then 1 = integer variable, 0 = continuous variable. <br>If any element ''>''1, IntVars is the indices for integer variables.
	\|-valign="top"		\|-valign="top"
	\|''VarWeight''\|\|Weight for each variable in the variable selection phase. <br> A lower value gives higher priority. Setting Prob.MIP.VarWeight might improve convergence.		\|''VarWeight''\|\|Weight for each variable in the variable selection phase. <br> A lower value gives higher priority. Setting Prob.MIP.VarWeight might improve convergence.<br>
			VarWeight must be of length N, but the values corresponding to noninteger variables will be ignored.
	\|-valign="top"		\|-valign="top"
	\|''DualGap''\|\|''stoaMINLP ''stops if the duality gap is less than DualGap. DualGap = 1, stop at first integer solution e.g. DualGap = 0.01, stop if solution ''< ''1% from optimal solution.		\|''DualGap''\|\|''stoaMINLP ''stops if the duality gap is less than DualGap. DualGap = 1, stop at first integer solution e.g. DualGap = 0.01, stop if solution ''< ''1% from optimal solution. <br>Note that a finite lower bound will only be set by the solver and thus generating a gap when Prob.Convex is set to true.<br>If Prob.Convex is set to false, the lower bound will be -Inf.
	\|-valign="top"		\|-valign="top"
	\|''fIP''\|\|An upper bound on the IP value wanted. Makes it possible to cut branches and avoid node computations. Used even if xIP not given.		\|''fIP''\|\|An upper bound on the IP value wanted. Makes it possible to cut branches and avoid node computations. Used even if xIP not given.
Line 161:		Line 162:
	\|''NodeSel''\|\|Node selection method: <br>= 1 Depth First. Priority on nodes with more integer components. Default <br>= 2 Breadth First. Priority on nodes with more integer components. <br>= 3 Pure LIFO (Last in, first out) Depth First. <br>= 4 Pure FIFO (First in, first out) Breadth First. <br>= 5 Maximize integer components. <br>= 6 Best bound.<br>= 7 Best estimate using pseudo-costs. <br>= 8 Best projection.		\|''NodeSel''\|\|Node selection method: <br>= 1 Depth First. Priority on nodes with more integer components. Default <br>= 2 Breadth First. Priority on nodes with more integer components. <br>= 3 Pure LIFO (Last in, first out) Depth First. <br>= 4 Pure FIFO (First in, first out) Breadth First. <br>= 5 Maximize integer components. <br>= 6 Best bound.<br>= 7 Best estimate using pseudo-costs. <br>= 8 Best projection.
	\|-valign="top"		\|-valign="top"
	\|''Backtrack''\|\|~~Node~~ selection method to ~~use after an integer solution has been found~~. <br>~~Available~~ choices ~~are the same as for~~ NodeSel. Default is 2, Breadth First.		\|''Backtrack''\|\|Switch node selection method during branch and bound.<br>
			When set to any of the NodeSel methods, node selection method is changed to the Backtrack value when the backtracking criterion (see parameter BackCrit below) is fulfilled.<br>For available choices, see NodeSel above. Default is 2, Breadth First.
			\|-valign="top"
			\|''BackCrit''\|\|Backtracking criterion, used when Backtrack > 0.<br>
			= 0. Integer solution found. Default.<br>
			= 1. Current node worse than best estimate of active nodes.
	\|-valign="top"		\|-valign="top"
	\|''VarSel''\|\|Variable selection method in branch and bound. <br>= 1 Use variable with most fractional value. <br>= 2 Use gradient and distance to nearest integer value. <br>= 3 Pseudo-cost-based branching.		\|''VarSel''\|\|Variable selection method in branch and bound. <br>= 1 Use variable with most fractional value. <br>= 2 Use gradient and distance to nearest integer value. <br>= 3 Pseudo-cost-based branching.

Bjorn: /* Prob structure */

2015-02-11T15:31:00Z

Prob structure

← Older revision		Revision as of 10:31, 11 February 2015
Line 125:		Line 125:
	If < 0, generate \|xInit\| points with Latin Hypercube design<br>		If < 0, generate \|xInit\| points with Latin Hypercube design<br>
	<br>		<br>
	~~dxm~~ matrix - Every column is an initial point (of length d = Prob.N)<br>		d x m matrix - Every column is an initial point (of length d = Prob.N)<br>
	<br>		<br>
	Default value for xInit is min(3000,max(60,N*6)) at all nodes except the root node<br>		Default value for xInit is min(3000,max(60,N*6)) at all nodes except the root node<br>

Bjorn: /* Prob structure */

2015-02-11T15:30:03Z

Prob structure

← Older revision		Revision as of 10:30, 11 February 2015
Line 111:		Line 111:
	\|-valign="top"		\|-valign="top"
	\|''SolverNLP''\|\|Name of the solver used for NLP subproblems. If empty, the default solver is found calling GetSolver('con',1); <br>If TOMLAB /SOL installed, SNOPT is the default solver.<br>If SolverNLP is a SOL solver (SNOPT, MINOS or NPSOL), the SOL.optPar and SOL.PrintFile is used:<br>See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters		\|''SolverNLP''\|\|Name of the solver used for NLP subproblems. If empty, the default solver is found calling GetSolver('con',1); <br>If TOMLAB /SOL installed, SNOPT is the default solver.<br>If SolverNLP is a SOL solver (SNOPT, MINOS or NPSOL), the SOL.optPar and SOL.PrintFile is used:<br>See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters
	~~\|-valign="top"~~
	\|''SolverLP''\|\|Name of the solver used for LP subproblems. If empty, the default solver is found calling GetSolver('lp',1). If TOMLAB /SOL or /SNOPT or /NPSOL installed, MINOS is the default solver. If SolverLP is a SOL solver (MINOS, SNOPT or NPSOL), the SOL.optPar and SOL.PrintFile is used: See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters.
	\|-		\|-
	\|''SolverNLP0''\|\|Name of the solver used for the initial NLP-problem. If empty, SolverNLP is used.		\|''SolverNLP0''\|\|Name of the solver used for the initial NLP-problem. If empty, SolverNLP is used.
	\|-valign="top"		\|-valign="top"
	\|''~~SolverUC~~''\|\|~~Solver~~ used for ~~the unconstrained feasibility problem, NLPF. NLPF is solved in case the NLP problem is infeasible~~. ~~<br>~~If empty, the default solver is found calling GetSolver('uc',~~LargeScale~~,0).		\|''SolverLP''\|\|Name of the solver used for LP subproblems. If empty, the default solver is found calling GetSolver('lp',1). If TOMLAB /SOL or /SNOPT or /NPSOL installed, MINOS is the default solver. If SolverLP is a SOL solver (MINOS, SNOPT or NPSOL), the SOL.optPar and SOL.PrintFile is used: See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters.
	\|-valign="top"		\|-valign="top"
	\|''RandState''\|\|If Convex == 0, RandState is sent to multiMin to initialize the random generator. RandState is used as follows: <br>If ''> ''0, rand('state',RandState) is set to initialize the pseudo-random generator.<br> If ''< ''0, rand('state',sum(100*clock)) is set to give a new set of random values each run. <br>If RandState == 0, rand('state',) is not called. <br>Default is RandState = -1		\|''RandState''\|\|If Convex == 0, RandState is sent to multiMin to initialize the random generator. RandState is used as follows: <br>If ''> ''0, rand('state',RandState) is set to initialize the pseudo-random generator.<br> If ''< ''0, rand('state',sum(100*clock)) is set to give a new set of random values each run. <br>If RandState == 0, rand('state',) is not called. <br>Default is RandState = -1
			\|-valign="top"
			\|''xInit''\|\|Parameter sent to the solver multiMin which is used to solve relaxed sub-problems for nonconvex problems (with Prob.Convex set to false). xInit determines the way initial points are generated and is set as follows:<br>
			<br>
			Either<br>
			1x1 number - <br>
			If > 0, the number of random initial points, default min(3000,30*n);<br>
			If < 0, generate \|xInit\| points with Latin Hypercube design<br>
			<br>
			dxm matrix - Every column is an initial point (of length d = Prob.N)<br>
			<br>
			Default value for xInit is min(3000,max(60,N*6)) at all nodes except the root node<br>
			\|-valign="top"
			\|''xInit0''\|\|Parameter sent to the solver multiMin when solving the initial relaxed NLP problem at the root node.<br>
			Default value for xInit0 is min(3000,max(100,N*10))<br>
			See xInit for more information.<br>
	\|-valign="top"		\|-valign="top"
	\|''MIP''\|\|Structure defining integer optimization parameters. See the table below for a description of each field used in ''Prob.MIP''.		\|''MIP''\|\|Structure defining integer optimization parameters. See the table below for a description of each field used in ''Prob.MIP''.

Bjorn: Created page with "{{Part Of Manual|title=the MIPNLP Manual|link=MIPNLP Manual}} ==Purpose== stoaMINLP solves convex mixed-integer nonlinear programming problems (MINLP). ''stoaMINLP '..."

2014-07-14T10:41:10Z

Created page with "{{Part Of Manual|title=the MIPNLP Manual|link=MIPNLP Manual}} ==Purpose== stoaMINLP solves convex mixed-integer nonlinear programming problems (MINLP). ''stoaMINLP '..."

New page

{{Part Of Manual|title=the MIPNLP Manual|link=[[MIPNLP|MIPNLP Manual]]}}
==Purpose==

stoaMINLP solves convex mixed-integer nonlinear programming problems (MINLP).

''stoaMINLP ''solves problems of the form

<math>
\begin{array}{cccccc}
\min\limits_{x} & f(x) \\s/t & x_{L} & \leq & x & \leq & x_{U} \\& b_{L} & \leq & Ax & \leq & b_{U} \\& c_{L} & \leq & c(x) & \leq & c_{U} \\& & & {x_{j} \in \mathbb{N} \ \ \forall j \in I} \\
\end{array}
</math>

where <math>x,x_{L},x_{U}\in \mathbb{R}^{n}</math> , <math>c(x),c_{L},c_{U}\in
\mathbb{R}^{m_{1}}</math> , <math>A\in \mathbb{R}^{m_{2}\times n}</math> and
<math>b_{L},b_{U}\in \mathbb{R}^{m_{2}}</math>. The variables <math>x \in I</math> , the
index subset of <math>1,...,n</math> are restricted to be integers.

==Calling Syntax==

Recommended is to first set IterPrint, to get information each iteration
<source lang="matlab">
Prob.optParam.IterPrint = 1;
</source>
Driver call, including printing with level 2:
<source lang="matlab">
Result = tomRun('stoaMINLP',Prob,2);
</source>
Direct solver call:
<source lang="matlab">
Result = stoaMINLP(Prob);
PrintResult(Result);
Result = tomRun('stoaMINLP',Prob,...)
</source>

=== Warm Start ===
To make a restart (warm start), just set the warm start flag, and call stoaMINLP once again:

<source lang="matlab">
Prob.WarmStart = 1;
Result = tomRun('stoaMINLP', Prob, 2);
</source>

stoaMINLP will read warm start information from the stoaMINLPSave.mat file. Another warm start (with same MaxFunc) is made by just calling tomRun again:

<source lang="matlab">
Result = tomRun('stoaMINLP', Prob, 2);
</source>

To make a restart from the warm start information in the Result structure, make a call to WarmDefGLOBAL before calling stoaMINLP. WarmDefGLOBAL moves information from the Result structure to the Prob structure and sets the warm start flag, <code>Prob.WarmStart = 1;</code>

<source lang="matlab">
Prob = WarmDefGLOBAL('stoaMINLP', Prob, Result);
</source>

where Result is the result structure returned by the previous run. A warm start (with same MaxIter) is done by just calling tomRun again:

<source lang="matlab">
Result = tomRun('stoaMINLP', Prob, 2);
</source>

To make another warm start with new MaxIter 100, say, redefine MaxIter as:

<source lang="matlab">
Prob.optParam.MaxIter = 100;
</source>

Then repeat the two lines:

<source lang="matlab">
Prob = WarmDefGLOBAL('stoaMINLP', Prob, Result);
Result = tomRun('stoaMINLP', Prob, 2);
</source>

==Inputs==

===Prob structure===
The '''TOMLAB''' problem structure

{| class="wikitable"
!colspan="2"|The following fields are used in ''Prob'':
|-
|'''Field'''||'''Description'''
|-
|''x_L''||Lower bounds on x.
|-
|''x_U''||Upper bounds on x.
|-
|''A''||The linear constraint matrix.
|-
|''b_L''||Lower bounds on linear constraints.
|-
|''b_U''||Upper bounds on linear constraints.
|-
|''c_L''||Lower bounds on nonlinear constraints.
|-
|''c_U''||Upper bounds on nonlinear constraints.
|-
|''x_0''||Starting point.
|-valign="top"
|''Convex''||If Convex==1, assume NLP problems are convex, and only one local NLP solver call is used at each node. If Convex==0 (Default), multiMin is used to do many calls to a local solver to determine the global minima at each node. The global minimum with most components integer valued is chosen.
|-valign="top"
|''MaxCPU''||Maximal CPU Time (in seconds) to be used by stoaMINLP, stops with best point found
|-valign="top"
|''PriLev''||Print level in stoaMINLP (default 1). Also see optParam.IterPrint
|-valign="top"
|''PriLevOpt''||Print level in sub-solvers (SNOPT and other NLP solvers): =0 No output; ''>''0 Convergence results. ''>''1 Output every iteration, ''>''2 Output each step in the NLP alg For other NLP solvers, see the documentation for the solver
|-valign="top"
|''WarmStart''||If true, ''>''0, stoaMINLP reads the output from the last run from Prob.stoaMINLP, if it exists. If it doesn't exist, stoaMINLP attempts to open and read warm start data from mat-file stoaMINLPSave.mat. stoaMINLP uses the warm start information to continue from the last run. The mat-file minlp- SolveSave.mat is saved every Prob.MIP.SaveFreq iteration.
|-valign="top"
|''SolverNLP''||Name of the solver used for NLP subproblems. If empty, the default solver is found calling GetSolver('con',1); If TOMLAB /SOL installed, SNOPT is the default solver. If SolverNLP is a SOL solver (SNOPT, MINOS or NPSOL), the SOL.optPar and SOL.PrintFile is used: See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters
|-valign="top"
|''SolverLP''||Name of the solver used for LP subproblems. If empty, the default solver is found calling GetSolver('lp',1). If TOMLAB /SOL or /SNOPT or /NPSOL installed, MINOS is the default solver. If SolverLP is a SOL solver (MINOS, SNOPT or NPSOL), the SOL.optPar and SOL.PrintFile is used: See help minosTL.m, npsolTL.m or snoptTL.m for how to set these parameters.
|-
|''SolverNLP0''||Name of the solver used for the initial NLP-problem. If empty, SolverNLP is used.
|-valign="top"
|''SolverUC''||Solver used for the unconstrained feasibility problem, NLPF. NLPF is solved in case the NLP problem is infeasible. If empty, the default solver is found calling GetSolver('uc',LargeScale,0).
|-valign="top"
|''RandState''||If Convex == 0, RandState is sent to multiMin to initialize the random generator. RandState is used as follows: If ''> ''0, rand('state',RandState) is set to initialize the pseudo-random generator. If ''< ''0, rand('state',sum(100*clock)) is set to give a new set of random values each run. If RandState == 0, rand('state',) is not called. Default is RandState = -1
|-valign="top"
|''MIP''||Structure defining integer optimization parameters. See the table below for a description of each field used in ''Prob.MIP''.
|-valign="top"
|''STOAMINLP''||Structure with solver-specific options. See the table below with descriptions of the fields used in ''Prob.STOAMINLP''.
|-valign="top"
|''optParam''||Structure with general optimization parameters. See the table below with descriptions of the fields used in ''Prob.optParam''.
|}

===MIP structure===

Substructure in Prob with parameters related to mixed integer programming.

{|class="wikitable"
!colspan="2"|fields used in '''Prob.MIP'''
|-
|'''Field'''||'''Description'''
|-valign="top"
|''IntVars''||If empty, all variables are assumed non-integer. If islogical(IntVars) (=all elements are 0/1), then 1 = integer variable, 0 = continuous variable. If any element ''>''1, IntVars is the indices for integer variables.
|-valign="top"
|''VarWeight''||Weight for each variable in the variable selection phase. A lower value gives higher priority. Setting Prob.MIP.VarWeight might improve convergence.
|-valign="top"
|''DualGap''||''stoaMINLP ''stops if the duality gap is less than DualGap. DualGap = 1, stop at first integer solution e.g. DualGap = 0.01, stop if solution ''< ''1% from optimal solution.
|-valign="top"
|''fIP''||An upper bound on the IP value wanted. Makes it possible to cut branches and avoid node computations. Used even if xIP not given.
|-valign="top"
|''xIP''||The x-values giving the fIP value, if a solution (xIP,fIP) is known.
|-valign="top"
|''NodeSel''||Node selection method: = 1 Depth First. Priority on nodes with more integer components. Default = 2 Breadth First. Priority on nodes with more integer components. = 3 Pure LIFO (Last in, first out) Depth First. = 4 Pure FIFO (First in, first out) Breadth First. = 5 Maximize integer components. = 6 Best bound. = 7 Best estimate using pseudo-costs. = 8 Best projection.
|-valign="top"
|''Backtrack''||Node selection method to use after an integer solution has been found. Available choices are the same as for NodeSel. Default is 2, Breadth First.
|-valign="top"
|''VarSel''||Variable selection method in branch and bound. = 1 Use variable with most fractional value. = 2 Use gradient and distance to nearest integer value. = 3 Pseudo-cost-based branching.
|-valign="top"
|''KNAPSACK''||If = 1, use a knapsack heuristic. Default 0.
|-valign="top"
|''ROUNDH''||If = 1, use a rounding heuristic. Default 0.
|-valign="top"
|''SaveFreq''||Warm start info saved on stoaMINLPSave.mat every SaveFreq iteration (default -1, i.e. no warm start info is saved)
|}

===STOAMINLP structure===

Substructure in Prob with solver-specific options

{|class="wikitable"
!colspan="2"|fields used in Prob.STOAMINLP
|-
|'''Field'''||'''Description'''
|-valign="top"
|''USECUTS''||Flag to use cutting plane algorithm described in [[StoaMINLP#Algorithm]]. Using cuts is not recommended for nonconvex problems. Set nonzero to enable, zero to disable. By default, USECUTS = Prob.Convex.
|-valign="top"
|''B_ecp''||Parameter regulating likelihood of linearization using extended cutting plane. See [[StoaMINLP#Algorithm]] for more information. Default 10.
|-valign="top"
|''K_ecp''||Parameter regulating "tailing off" detection for extended cutting plane linearizations. See [[StoaMINLP#Algorithm]] for more information. Default 0.001.
|-valign="top"
|''M_ecp''||Parameter regulating the balance between branching the search tree and adding linearizations using extended cutting plane. See [[StoaMINLP#Algorithm]] for more information. Default 10.
|-valign="top"
|''B_ff''||Parameter regulating likelihood of linearization using fixed fractional integer decision variables. See [[StoaMINLP#Algorithm]] for more information. Default 10.
|-valign="top"
|''K_ff''||Parameter regulating "tailing off" detection for linearizations using fixed fractional integer decision variables. See [[StoaMINLP#Algorithm]] for more information. Default 0.001.
|-valign="top"
|''M_ff''||Parameter regulating the balance between branching the search tree and adding linearizations using fixed fractional integer decision variables. See [[StoaMINLP#Algorithm]] for more information. Default 10.
|-valign="top"
|''B_nlpr''||Parameter regulating likelihood of linearization using the solution of the relaxed NLP problem. See [[StoaMINLP#Algorithm]] for more information. Default 10.
|-valign="top"
|''K_nlpr''||Parameter regulating "tailing off" detection for linearizations using the solution of the relaxed NLP problem. See [[StoaMINLP#Algorithm]] for more information. Default 0.001.
|-valign="top"
|''M_nlpr''||Parameter regulating the balance between branching the search tree and adding linearizations using the solution of the relaxed NLP problem. See [[StoaMINLP#Algorithm]] for more information. Default 10.
|-valign="top"
|''rTYPE''||= 0 r is fixed to the value in rVALUE. = 1 r is a uniformly distributed random number in the interval [0,1]. Default.
|-valign="top"
|''rVALUE''||Defines the value of r when rTYPE is set to zero. Default is 0.5.
|-valign="top"
|''CUTPOOL''||= 0 All linearizations constraints are kept in the master problem. = 1 Manage a pool with inactive linearization constraints. If an inactive constraint gets violated, it is brought back from the pool. Default.
|-valign="top"
|''CUTDELBND''||The number of consecutive times a linearization can be inactive before it gets removed and put into the inactive pool. Default 15.
|-valign="top"
|''CUTLIMDEL''||= 0 No extra deletion of cuts. Default when Prob.Convex == 1. = 1 Delete any cuts limiting the true feasible region when detected. Such cuts will appear if the problem has nonlinear constraints. Default when Prob.Convex == 0.
|-valign="top"
|''PCITL''||Limit on time spent per node calculating initial pseudo costs when VarSel == 3. Default is 120 (s).
|-valign="top"
|''LINPRINT''||Set nonzero to print short information about the type of linearization added. Default 0. enabled when IterPrint > 3.
|-valign="top"
|''POOLPRINT''||Set nonzero to print information about the cutpool management. Default 0. Enabled when IterPrint > 4.
|}

===optParam structure===

Substructure in Prob with general optimization parameters.

{|class="wikitable"
!colspan="2"|fields used in Prob.optParam:
|-
|'''Field'''||'''Description'''
|-valign="top"
|''MaxIter''||Maximal number of iterations, default 10000
|-valign="top"
|''IterPrint''||Print short information each iteration. Setting PriLev > 1 also enables IterPrint, setting it to 1. 
The information printed for different levels is as follows: 

>0 A line at each evaluated node with the following: 
Iteration number: 
Depth in tree (symbol L[] - empty list, symbol Gap - Dual Gap convergence), fNLP (Optimal f(x) current node), fIPMin (Best integer feasible f(x) found), LowBnd (Lower bound on optimal integer feasible f(x)), Dual Gap in absolut value and percent, The length of the node list L, -L-, The Inform and ExitFlag the solver returned at the current node, FuEv (Number of function evaluations used by solver at current node), date/time stamp. Outer approximation linearizations, number of cuts in active/inactive pool (|C| #A #I). 

> 1 A line at each evaluated node with the information described above. 
Moreover, with IterPrint enabled a notice is printed whenever the algorithm switches node selection methods. 
Setting a higher value of IterPrint gives more detailed information. 

>2 Depth, number of integer components and lower bound on objective for each node in L. 
>3 Sets LINPRINT = 1, print info about linearizations added (if used). 
>4 Sets POOLPRINT = 1, print info about cut-pool management (if used).
|-valign="top"
|''bTol''||Linear constraint violation convergence tolerance.
|-valign="top"
|''cTol''||Constraint violation convergence tolerance.
|-valign="top"
|''fTol''||Optimality tolerance.
|}

==Outputs==

{| class="wikitable"
!''Result''||colspan="2"|Structure with result from optimization. The following fields are changed:
|-
|||''Iter ''||Number of iterations.
|-valign="top"
|||''ExitFlag ''||Exit flag. 0: Global optimal solution found, or integer solution with duality gap less than user tolerance. 1: Maximal number of iterations reached. 2: Empty feasible set, no integer solution found. 4: No feasible point found running NLP relaxation. 5: Illegal x_0 found in NLP relaxation. 99: Maximal CPU Time used (cputime ''> ''Prob.MaxCPU).
|-
|||''Inform''||Code telling type of convergence, returned from subsolver.
|-
|||''ExitText''||Text string giving ''ExitFlag'' and ''Inform'' information.
|-valign="top"
|||''DualGap''||Relative duality gap, max(0,fIPMin-fLB)/-fIPMin-, if fIPMin =0; max(0,fIPMin-fLB) if fIPMin == 0. If fIPMin =0: Scale with 100, 100*Dual- Gap, to get the percentage duality gap. For absolute value duality gap: scale with fIPMin, fIPMin * DualGap
|-
|||''x_k''||Solution.
|-valign="top"
|||''v_k''||Lagrange multipliers. Bounds, Linear and Nonlinear Constraints, size n + mLin + mNonLin.
|-
|||''f_k''||Function value at optimum.
|-
|||''g_k''||Gradient vector at optimum.
|-
|||''x_0''||Starting point x_0.
|-
|||''f_0''||Function value at start.
|-
|||''c_k''||Constraint values at optimum.
|-
|||''cJac''||Constraint derivative values at optimum.
|-
|||''xState''||State of each variable, described in [[TOMLAB_Appendix_B#Table:_The_state_variable_xState_for_the_variable.|TOMLAB Appendix B]].
|-
|||''bState''||State of each constraint, described in [[TOMLAB_Appendix_B#Table:_The_state_variable_bState_for_each_linear_constraint.|TOMLAB Appendix B]].
|-
|||''cState''||State of each general constraint, described in [[TOMLAB_Appendix_B#Table:_The_state_variable_cState_for_each_nonlinear_constraint.|TOMLAB Appendix B]].
|-
|||''Solver''||Solver used ('stoaMINLP').
|-
|||''SolverAlgorithm''||Description of method used.
|-
|||''Prob''||Problem structure used.
|-valign="top"
|||''stoaMINLP''||A structure with warm start information. Use with WarmDefGLOBAL, see example [[StoaMINLP#Warm_Start|here]].
|}

[[Category:MIPNLP]]