[breaking] Configuration templates for the BundleSolver 2.0

docs/release_notes.md

@@ -8,7 +8,11 @@
 
 ### Minor Changes and Bug Fixes
 
-*
+* Write the templates that use the BundleSolver for the BundleSolver 2.0, whose master is configured by the new `MPBCfg.txt` (HiGHS) or `MPBCfg_grb.txt` (Gurobi). See [PR #116](https://github.com/SPSUnipi/pySMSpp/pull/116)
+
+* Add templates for the two-stage instances whose scenario is a unit commitment: the MILP (`TSSBSCfg-IP.txt`), the Lagrangian dual of the scenarios (`TSSBSCfg-LD-IP.txt`), the chain of two duals (`TSSBSCfg-LDLD.txt`), the recursive Lagrangian dual (`TSSBSCfg-LDrec.txt`) and the PrimalProximalHeur, which gives a feasible solution (`TSSBSCfg-PPH.txt`), the three duals also with a ParallelBundleSolver (`-par`), and the continuous relaxation with Gurobi (`TSSBSCfg-LP.txt`) and with PIPS-IPM++ (`TSSBSCfg-PIPS.txt`). See [PR #116](https://github.com/SPSUnipi/pySMSpp/pull/116)
+
+* Add the templates of the BendersDecompositionSolver for the Benders form of a two-stage problem (`TSSBSCfg-BDS.txt`, `BDSMCfg.txt`, `BDSSCfg.txt`), and of the InvestmentBlock over a TSSB or an MSSB (`TSSBSCfg-IB.txt`, `BSCfg-IB.txt`). The Solver of the block inside an InvestmentBlock is given by its OBlockConfig (`IBOCfg.txt`, `IFCfg.txt`) through the `InvestmentBlock` entry of `InnerBCfg.txt`, which `smspp_investmentblock_solver` reads by default, in `InvestmentBlock/` and in `TSSBlock/`. See [PR #116](https://github.com/SPSUnipi/pySMSpp/pull/116)
 
 
 ## Version v0.0.14

pysmspp/data/blocks/InvestmentBlock.csv

@@ -10,6 +10,7 @@ InvestmentBlock,DisinvestmentCost,Variable,Variable,double,-|NumAssets,-,,requir
 InvestmentBlock,LowerBound,Variable,Variable,double,-|NumAssets,-,,optional,Minimum capacity of each asset
 InvestmentBlock,UpperBound,Variable,Variable,double,-|NumAssets,-,,optional,Maximum capacity of each asset
 InvestmentBlock,InstalledQuantity,Variable,Variable,double,-|NumAssets,-,,optional,Installed capacity of each asset
+InvestmentBlock,Integer,Variable,Variable,int,-|NumAssets,-,,optional,Nonzero if the design of the asset is integer (e.g. the number of its modules)
 InvestmentBlock,Constraints_LowerBound,Variable,Variable,double,-|NumConstraints,-,,optional,Lower bound of each constraint
 InvestmentBlock,Constraints_UpperBound,Variable,Variable,double,-|NumConstraints,-,,optional,Upper bound of each constraint
 InvestmentBlock,Constraints_A,Variable,Variable,double,-|NumConstraints-NumAssets,-,,optional,Coefficients of matrix of the linear constraints

pysmspp/data/configs/InvestmentBlock/BSCfg_grb.txt

@@ -3,13 +3,11 @@
 #
 # Gurobi variant of BSCfg.txt: a BlockSolverConfig for the inner UCBlock-style
 # block of an InvestmentBlock, to be solved by GRBMILPSolver as the LP
-# relaxation (intRelaxIntVars 1) of the underlying MIP. Presolve is disabled
-# for reproducibility (it can transform the model and shift the optimal vertex
-# on degenerate LPs), and NumericFocus is raised because the InvestmentBlock
-# inner LPs are badly scaled (capacities, snapshot-weighted costs and VOLL span
-# many orders of magnitude), which makes HiGHS fail on some iterates.
-#
-#   Antonio Frangioni, Donato Meoli
+# relaxation (intRelaxIntVars 1) of the underlying MIP, with the presolve and
+# the numerical focus of Gurobi left at their default: the inner LP is solved
+# at every evaluation of the InvestmentFunction, and with Presolve 0 and
+# NumericFocus 3 each evaluation cost 2.5 times as much, which the badly
+# scaled LPs that make HiGHS fail (see BSCfg.txt) do not need with Gurobi.
 #
 # - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 # BlockSolverConfig - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
@@ -33,16 +31,14 @@ ComputeConfig # exact type of the ComputeConfig object
 1  # f_diff == 0 ==> all non-provided parameters are set to the default value
    # f_diff == 1 ==> all non-provided parameters are not changed
 
-5  # number of integer parameters
+3  # number of integer parameters
 
 # now all the integer parameters
 intLogVerb          0    # LogVerb, log verbosity of the MILPSolver
 intRelaxIntVars     1    # nonzero if the continuous relaxation is solved
 intCutSepPar        7    # separate user cuts + lazy constraints
 
 # All specific GUROBI parameter
-Presolve      0    # disable presolve for reproducibility
-NumericFocus  3    # maximum numerical precision focus
 
 0  # number of double parameters
 

pysmspp/data/configs/InvestmentBlock/BSPar-int.txt

@@ -0,0 +1,72 @@
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - - BSPar-int.txt - - - - - - - - - - - - - - -
+#
+# A BlockSolverConfig for an InvestmentBlock whose design is integer, e.g.,
+# the number of modules of a modular asset: the BundleSolver of BSPar.txt
+# keeping it integer, with a trust region of radius 1 and a master that is a
+# mixed-integer linear problem [see intIntVars].
+#
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+BlockSolverConfig     # exact type of the Configuration object
+
+1  # the BlockSolverConfig is a "differential" one
+
+1  # number of (the names of) Solver in this BlockSolverConfig
+# now all the names of the Solver - - - - - - - - - - - - - - - - - - - - - -
+BundleSolver    # name of Solver
+
+1  # number of ComputeConfig in this BlockSolverConfig
+
+# now all the ComputeConfig
+# 1st ComputeConfig - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+# ComputeConfig of the BundleSolver
+
+*BundleCfg.txt +
+
+1  # f_diff == 1 ==> only the entries below are changed
+
+5  # number of integer parameters
+
+# now all the integer parameters
+intIntVars   1  # the integer design, e.g., the modules, stay integer
+intMPPrimal  1  # the primal master, where the design is the Variable
+intMPV2Form  1  # raw form: the Variable of the master are the design itself
+intMPStbl    4  # 4 = trust region || x - x_bar ||_inf <= t, a MILP master
+intRstAlg    2  # parameter to handle the reset of the algorithm
+
+1  # number of double parameters
+
+# now all the double parameters
+dbltInit     1  # radius of the trust region, in modules
+
+1  # number of string parameters
+
+# now all the string parameters
+strMPBSolverCfg  MPBCfg-IP.txt  # the master, which keeps the integers
+
+0  # number of vector-of-int parameters
+
+# now all the vector-of-int parameters
+# [none]
+
+0  # number of vector-of-double parameters
+
+# now all the vector-of-double parameters
+# [none]
+
+0  # number of vector-of-string parameters
+
+# now all the vector-of-string parameters
+# [none]
+
+# pointer to the "extra" Configuration
+* # [none]
+
+# end of 1st ComputeConfig- - - - - - - - - - - - - - - - - - - - - - - - - -
+
+# end of BlockSolverConfig- - - - - - - - - - - - - - - - - - - - - - - - - -
+
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - END BSPar-int.txt - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

pysmspp/data/configs/InvestmentBlock/BSPar.txt

@@ -1,12 +1,9 @@
 # - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
-# - - - - - - - - - - - - - - - - BSPar.txt - - - - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - - - BSPar.txt - - - - - - - - - - - - - - - -
 #
-# A txt description of a BlockSolverConfig for an InvestmentBlock to be
-# solved by a BundleSolver
+# A BlockSolverConfig for an InvestmentBlock to be solved by a
+# BundleSolver [see BundleCfg.txt].
 #
-#   Antonio Frangioni
-#
-# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 # - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
 
 BlockSolverConfig     # exact type of the Configuration object
@@ -21,93 +18,9 @@ BundleSolver    # name of Solver
 
 # now all the ComputeConfig
 # 1st ComputeConfig - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
-# ComputeConfig of the LagrangianDualSolver; it mostly is the
-# ComputeConfig of the inner Solver, which is a [Parallel]BundleSolver
-
-ComputeConfig # exact type of the ComputeConfig object
-
-1  # f_diff == 0 ==> all non-provided parameters are set to the default value
-   # f_diff == 1 ==> all non-provided parameters are not changed
-
-23 # number of integer parameters
-
-# now all the integer parameters
-intDoEasy  0      # do easy components for (some) LagBFunction
-intMaxIter 10000  # max number of iterations for each call
-intMaxThread 6	   # MaxThread, max number of new tasks to spawn
-intLogVerb 16      # log verbosity of main Bundle algorithm
-intWZNorm   2     # which norm to use in the norm-based stopping condition
-intBPar1   20     # discard items when they have been useless for <this> iterations
-intBPar2 1000     # max bundle size per component
-intBPar3    1     # max n. of items to fetch from Fi() at each iteration
-intBPar4    1     # min n. of items to fetch from Fi() at each iteration
-intBPar6    0     # second parameter for dynamic max n. of items per iteration
-intBPar7   15     # how to deal with the global pools
-intMnSSC    0     # min number of consecutive SS with the same t for a t increase
-intMnNSC    3     # min number of consecutive NS with the same t for a t decrease
-inttSPar1   0     # long-term t-strategy (0 = none, 4 = soft, 8 = hard, 12 = balancing)
-intMaxNrEvls 2    # maximum number of function evaluations at each iteration
-intMPName  8      # MP solver: 0 = QPP, 7 = Cplex/Gurobi with quadratic stabilization
-                  # + 8 = check for duplicate linearizations
-
-# QPPenalty's parameters :- - - - - - - - - - - - - - - - - - - - - - - - - -
-
-intMPlvl   20     # log verbosity of Master Problem
-intQPmp1   0	  # MxAdd, how many variables can be added to the base at each
-      	          # iteration in BMinQuad (0 = at will)
-intQPmp2   0      # MxRmv, how many variables can be removed from the base at each
-      	          # iteration in BMinQuad (0 = at will)
-
-# OSIMPSolver's parameters :- - - - - - - - - - - - - - - - - - - - - - - - -
-
-intOSImp1  1      # Osi(Cpx/Grb)Solver algorithm (0 = auto (default), 1 = primal, 2 = dual, 3 = network, 4 = barrier, 5 = sifting, 6 = concurrent)
-intOSImp2  1      # pre-processing (reduction)
-intOSImp3  1      # threads
-intRstAlg  2      # parameter to handle the reset of the algorithm
-
-16 # number of double parameters
-
-# now all the double parameters
-dblRelAcc  1e-4   # relative accuracy required to solution
-dblNZEps   1e-4   # stopping parameter: threshold to declare 0 the || residual ||
-dbltStar   -100   # stopping parameter: multiplied to || residual || to estimate gap
-dblBPar5   4      # first parameter for dynamic max n. of items per iteration
-dblm1      -0.30  # a NS is possible if (~) Fi( Lambda1 ) >= Fi( Lambda ) + | m1 | v*
-dblm2      0.90   # a SS is possible if Fi( Lambda1 ) <= Fi( Lambda ) + ( 1 - m2 ) v*
-dblm3      0.99   # a NR is computed if \sigma^* < - t * m3 * || z* ||
-dblmxIncr  10     # max increase of t
-dblmnIncr  1.5    # min increase of t (each time it is increased)
-dblmxDecr  0.1    # max decrease of t
-dblmnDecr  0.66   # min decrease of t (each time it is decreased)
-dbltMaior  1e+6   # maximum value for t
-dbltMinor  1e-10  # minimum value for t
-dbltInit   1e+0   # initial value for t
-dbltSPar2  1e-2   # parameter for the long-term t-strategy
-dblCtOff   0.01   # cut-off factor for pricing in MinQuad
-
-0 # number of string parameters
-
-# now all the string parameters
-#str_LagBF_BSCfg	LPBSCfg.txt   # BlockSolverConfig for the LagBFunctions
-# not needed if intDoEasy >= 1 && USE_BundleSolver > 0, this is done "by hand"
-
-0 # number of vector-of-int parameters
-
-# now all the vector-of-int parameters
-# [none]
-
-0 # number of vector-of-double parameters
-
-# now all the vector-of-double parameters
-# [none]
-
-0 # number of vector-of-string parameters
-
-# now all the vector-of-string parameters
-# [none]
+# ComputeConfig of the BundleSolver
 
-# pointer to the "extra" Configuration
-* # [none]
+*BundleCfg.txt
 
 # end of 1st ComputeConfig- - - - - - - - - - - - - - - - - - - - - - - - - -
 

pysmspp/data/configs/InvestmentBlock/BundleCfg.txt

@@ -0,0 +1,82 @@
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - - BundleCfg.txt - - - - - - - - - - - - - - -
+#
+# The ComputeConfig of the BundleSolver that solves an InvestmentBlock,
+# shared by BSPar.txt and BSPar-int.txt.
+#
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+ComputeConfig # exact type of the ComputeConfig object
+
+1  # f_diff == 0 ==> all non-provided parameters are set to the default value
+   # f_diff == 1 ==> all non-provided parameters are not changed
+
+17 # number of integer parameters
+
+# now all the integer parameters
+intMPStbl              0 # stabilization: 0 = proximal, 1 = level, 2 = doubly
+intDoEasy  0      # do easy components for (some) LagBFunction
+intMaxIter 10000  # max number of iterations for each call
+intMaxThread 6	   # MaxThread, max number of new tasks to spawn
+intLogVerb 16      # log verbosity of main Bundle algorithm
+intWZNorm   2     # which norm to use in the norm-based stopping condition
+intBPar1   20     # discard items when they have been useless for <this>
+intBPar2 1000     # max bundle size per component
+intBPar3    1     # max n. of items to fetch from Fi() at each iteration
+intBPar4    1     # min n. of items to fetch from Fi() at each iteration
+intBPar6    0     # second parameter for dynamic max n. of items per
+intBPar7   15     # how to deal with the global pools
+intMnSSC    0     # min number of consecutive SS with the same t for a t
+intMnNSC    3     # min number of consecutive NS with the same t for a t
+inttSPar1  12     # long-term t-strategy (0 = none, 4 = soft, 8 = hard, 12 =
+intMaxNrEvls 2    # maximum number of function evaluations at each iteration
+
+intRstAlg  2      # parameter to handle the reset of the algorithm
+
+15 # number of double parameters
+
+# now all the double parameters
+dblRelAcc  1e-4   # relative accuracy required to solution
+dblNZEps   1e-4   # stopping parameter: threshold to declare 0 the ||
+dbltStar   -100   # stopping parameter: multiplied to || residual || to
+dblBPar5   4      # first parameter for dynamic max n. of items per iteration
+dblm1      -0.30  # a NS is possible if (~) Fi( Lambda1 ) >= Fi( Lambda ) + |
+dblm2      0.90   # a SS is possible if Fi( Lambda1 ) <= Fi( Lambda ) + ( 1 -
+dblm3      0.99   # a NR is computed if \sigma^* < - t * m3 * || z* ||
+dblmxIncr  10     # max increase of t
+dblmnIncr  1.5    # min increase of t (each time it is increased)
+dblmxDecr  0.1    # max decrease of t
+dblmnDecr  0.66   # min decrease of t (each time it is decreased)
+dbltMaior  1e+6   # maximum value for t
+dbltMinor  1e-10  # minimum value for t
+dbltInit   1e+0   # initial value for t
+dbltSPar2  1e-2   # parameter for the long-term t-strategy
+
+1 # number of string parameters
+
+# now all the string parameters
+#str_LagBF_BSCfg	LPBSCfg.txt   # BlockSolverConfig for the LagBFunctions
+# not needed if intDoEasy >= 1 && USE_BundleSolver > 0, this is done "by
+strMPBSolverCfg   MPBCfg.txt # BlockSolverConfig for the MasterProblemBlock's
+
+0 # number of vector-of-int parameters
+
+# now all the vector-of-int parameters
+# [none]
+
+0 # number of vector-of-double parameters
+
+# now all the vector-of-double parameters
+# [none]
+
+0 # number of vector-of-string parameters
+
+# now all the vector-of-string parameters
+# [none]
+
+# pointer to the "extra" Configuration
+* # [none]
+
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - END BundleCfg.txt - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

pysmspp/data/configs/InvestmentBlock/IBOCfg.txt

@@ -0,0 +1,54 @@
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - - - IBOCfg.txt  - - - - - - - - - - - - - - - -
+#
+# The OBlockConfig of an InvestmentBlock: it reformulates the bounds on the
+# investment, and gives the InvestmentFunction its ComputeConfig [see
+# IFCfg.txt], which holds the BlockSolverConfig of the inner Block.
+#
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+
+OBlockConfig  # exact type of the Configuration object
+
+1  # the BlockConfig is a "differential" one
+
+2  # version of the BlockConfig format
+
+# the Configuration for the structure, i.e., the tree of sub-Block
+*  # [none]
+
+# static_constraints: the bounds l <= x <= u on the investment are
+# reformulated as 0 <= x <= u - l, the only form BundleSolver handles
+SimpleConfiguration<int>
+1
+
+# dynamic_constraints
+*  # [none]
+
+# static_variables
+*  # [none]
+
+# dynamic_variables
+*  # [none]
+
+# objective
+*  # [none]
+
+# is_feasible
+*  # [none]
+
+# is_optimal
+*  # [none]
+
+# solution
+*  # [none]
+
+# extra
+*  # [none]
+
+# the ComputeConfig of the Objective, i.e., of the InvestmentFunction
+*IFCfg.txt
+
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - - END IBOCfg.txt  - - - - - - - - - - - - - - -
+# - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -