MiniZinc - Challenge 2016 Results

MiniZinc Challenge 2016 Results

CP 2016 presentation slides of the MiniZinc Challenge 2016 (results).

Entrants

The entrants for this year (with their descriptions, when provided):

Choco 3 (description). A Java FD solver.
Concrete (description). A CP system written in Scala.
HaifaCSP (description). A C++ FD solver using SAT solving algorithms.
iZplus (description).
JaCoP (description). A Java FD solver.
MinisatID (description). A solver combining techniques from the fields SAT, SMT, CP, and ASP.
Mistral 2.0 (description). A C++ FD solver.
OR-Tools (description).
OscaR/CBLS (description). A constraint-based local search solver written in Scala.
Picat CP (description).
Picat SAT (description).
SICStus Prolog (description). A Prolog development environment with a FD constraint programming module.
sunny-cp^— (see sunny-cp for description). A variant of sunny-cp only using Choco, Gecode, HaifaCSP, JaCoP, iZplus, MinisatID, Mistral, Opturion CPX, OR-Tools, Picat.
sunny-cp^—— (see sunny-cp for description). A variant of sunny-cp only using Gecode, HaifaCSP, JaCoP, MinisatID, Mistral, Picat.
Yuck (description). A local search solver written in Scala.

In addition, the challenge organisers entered the following FlatZinc and MiniZinc implementations:

Chuffed (description). A C++ FD solver using lazy clause generation.
G12/FD. A Mercury FD solver (the G12 FlatZinc interpreter's default solver).
Gecode (description). A C++ FD solver.
LCG-Glucose (description). A C++ FD solver using lazy clause generation and it is based on Glucose 3.01 with modification along the lines of Chuffed and CPX.
LCG-Glucose-UC (description). A variant of LCG-Glucose using unsat cores.
MZN/Cbc. Translates to MILP (description), uses Cbc version 2.9.8.
MZN/CPLEX. Translates to MILP (description), uses IBM ILOG CPLEX Optimizer version 12.6.3.
MZN/Gurobi. Translates to MILP (description), uses Gurobi Optimizer version 6.5.2.
MZN/SCIP. Translates to MILP, uses SCIP Optimization Suite version 3.2.1.
sunny-cp (description). A multi-threaded CP portfolio solver using 13 different CP and MIP solvers incl. Chuffed, G12 solvers, Gecode.

As per the challenge rules, these entries are not eligible for prizes, but do modify the scoring results. Furthermore, entries in the FD search category (JaCoP, Picat CP, SICStus Prolog) were automatically included in the free search category, while entries in the free search category (Chuffed, Concrete, G12/FD, LCG-Glucose, LCG-Glucose-UC, MZN/Cbc, MZN/SCIP, OscaR/CBLS, Picat SAT and promoted FD entries) were automatically included in the parallel search category. Lastly, all entries in the parallel search category and promoted entries into that category were automatically included in the open search category.

Summary of Results

The results for the MiniZinc Challenge 2016 are

Category	Gold	Silver	Bronze
Fixed	OR-Tools	JaCoP	Choco
Free	HaifaCSP	Picat SAT	iZplus
Parallel	HaifaCSP	Picat SAT	iZplus
Open	sunny-cp——	HaifaCSP	Picat SAT
Local Search	OscaR/CBLS

Description of Results

All times are given in milliseconds.

A score of 0.0 indicates a worse answer in quality (worse objective, no proof of optimality, or no answer for satisfaction problems), 1.0 a better solution in quality. When the quality is the same, the 1.0 purse is split with respect to time used.

If a promoted entry does not recognize an option (or states that it is just ignored), times and solutions from the previous category are used for scoring. The suffixes -fd, -free, -par or -open (for the parallel portfolio solver entered) at the end of the solver names indicate which configuration the solvers were run with.

mzn2fzn was run with the same time and memory limits as the solvers.

In the Status column:

S indicates that a solution was found,
C indicates that the search was complete,
UNK indicates an incorrect answer
UNK indicates that flattening aborted (time-out or out-of-memory)
UNK indicates that no answer was returned in the time limit or the solver aborted.

Download all problems

All problems are available in a zipped tar-ball here.

Selection:

Select a list of solvers and benchmarks and click on “Compute Results” to score the solvers against each other on the selected benchmarks. The entrants for each of the fd search, free search and parallel search categories can be selected with the corresponding buttons.

Solver selection:	Problem selection:

Summary:

Total per problem:

Solver	Score	Score Incomplete

Problem	Solver	Score	Score Incomplete

Individual results:

Problem	Instance	Solver	Status	Time	Objective	Score	Score Incomplete

Global constraint per model

The following table lists the global constraints used by each model in this year's challenge. In addition, the columns RC and SBC, respectively, indicate whether the model contains redundant or/and symmetry breaking constraints.

Problem	Type	Kind	RC	SBC	MiniZinc Globals
carpet-cutting	real	min		✓	cumulative, diffn
celar	combi	min
cryptanalysis	real	sat
depot_placement	combi	min		✓	all_different
diameterc-mst	combi	min	✓	✓
elitserien	real	min			all_different, global_cardinality_closed, inverse, member, regular
filters	combi	min			diffn, maximum
gbac	combi	min			bin_packing_load, global_cardinality_low_up_closed
gfd-schedule2	real	min		✓	at_most, cumulative, nvalue
java-auto-gen	combi	min			all_different, element
mapping	real	min			bin_packing_load, count, maximum, network_flow_cost
maximum-dag	combi	max
mrcpsp	combi	min	✓		cumulative, element
nfc	combi	min			network_flow_cost
oocsp_racks	real	sat			global_cardinality, increasing
prize-collecting	puzzle	max
rcpsp-wet	combi	min	✓		cumulative
solbat	combi	sat
tpp	combi	min			subcircuit
zephyrus-2016	real	min	✓	✓	lex_greatereq, lex_less

The files on this page are for MiniZinc version 2.0.13.