solve_TSP: TSP solver interface

Description Usage Arguments Details Value Author(s) References See Also Examples

View source: R/solve_TSP.R

Description

Common interface to all TSP solvers in this package.

Usage

1
solve_TSP(x, method = NULL, control = NULL, ...)

Arguments

x

the TSP given as an object of class TSP, ATSP or ETSP.

method

method to solve the TSP (default: arbitrary insertion algorithm with two_opt refinement.

control

a list of arguments passed on to the TSP solver selected by method.

...

additional arguments are added to control.

Details

Treatment of NAs and infinite values in x: TSP and ATSP contain distances and NAs are not allowed. Inf is allowed and can be used to model the missing edges in incomplete graphs (i.e., the distance between the two objects is infinite). Internally, Inf is replaced by a large value given by max(x) + 2 range(x). Note that the solution might still place the two objects next to each other (e.g., if x contains several unconnected subgraphs) which results in a path length of Inf.

All heuristics can be used with the control arguments repetitions (uses the best from that many repetitions with random starts) and two_opt (a logical indicating if two_opt refinement should be performed). If several repetitions are done (this includes method "repetitive_nn") then foreach is used so they can be performed in parallel on multiple cores/machines. To enable parallel execution an appropriate parallel backend needs to be registered (e.g., load doParallel and register it with registerDoParallel()).

ETSP are currently solved by first calculating a dissimilarity matrix (a TSP). Only concorde and linkern can solve the TSP directly on the ETSP.

Currently the following methods are available:

"identity", "random"

return a tour representing the order in the data (identity order) or a random order.

"nearest_insertion", "farthest_insertion", "cheapest_insertion", "arbitrary_insertion"

Nearest, farthest, cheapest and arbitrary insertion algorithms for a symmetric and asymmetric TSP (Rosenkrantz et al. 1977).

The distances between cities are stored in a distance matrix D with elements d(i,j). All insertion algorithms start with a tour consisting of an arbitrary city and choose in each step a city k not yet on the tour. This city is inserted into the existing tour between two consecutive cities i and j, such that

d(i,k) + d(k,j) - d(i,j)

is minimized. The algorithms stops when all cities are on the tour.

The nearest insertion algorithm chooses city k in each step as the city which is nearest to a city on the tour.

For farthest insertion, the city k is chosen in each step as the city which is farthest to any city on the tour.

Cheapest insertion chooses the city k such that the cost of inserting the new city (i.e., the increase in the tour's length) is minimal.

Arbitrary insertion chooses the city k randomly from all cities not yet on the tour.

Nearest and cheapest insertion tries to build the tour using cities which fit well into the partial tour constructed so far. The idea behind behind farthest insertion is to link cities far away into the tour fist to establish an outline of the whole tour early.

Additional control options:

start

index of the first city (default: random city).

"nn", "repetitive_nn"

Nearest neighbor and repetitive nearest neighbor algorithms for symmetric and asymmetric TSPs (Rosenkrantz et al. 1977).

The algorithm starts with a tour containing a random city. Then the algorithm always adds to the last city on the tour the nearest not yet visited city. The algorithm stops when all cities are on the tour.

Repetitive nearest neighbor constructs a nearest neighbor tour for each city as the starting point and returns the shortest tour found.

Additional control options:

start

index of the first city (default: random city).

"two_opt"

Two edge exchange improvement procedure (Croes 1958).

This is a tour refinement procedure which systematically exchanges two edges in the graph represented by the distance matrix till no improvements are possible. Exchanging two edges is equal to reversing part of the tour. The resulting tour is called 2-optimal.

This method can be applied to tours created by other methods or used as its own method. In this case improvement starts with a random tour.

Additional control options:

tour

an existing tour which should be improved. If no tour is given, a random tour is used.

two_opt_repetitions

number of times to try two_opt with a different initial random tour (default: 1).

"concorde"

Concorde algorithm (Applegate et al. 2001).

Concorde is an advanced exact TSP solver for only symmetric TSPs based on branch-and-cut. The program is not included in this package and has to be obtained and installed separately (see Concorde).

Additional control options:

exe

a character string containing the path to the executable (see Concorde).

clo

a character string containing command line options for Concorde, e.g., control = list(clo = "-B -v"). See concorde_help on how to obtain a complete list of available command line options.

precision

an integer which controls the number of decimal places used for the internal representation of distances in Concorde. The values given in x are multiplied by 10^{precision} before being passed on to Concorde. Note that therefore the results produced by Concorde (especially lower and upper bounds) need to be divided by 10^{precision} (i.e., the decimal point has to be shifted precision placed to the left). The interface to Concorde uses write_TSPLIB (see there for more information).

"linkern"

Concorde's Chained Lin-Kernighan heuristic (Applegate et al. 2003).

The Lin-Kernighan (Lin and Kernighan 1973) heuristic uses variable k edge exchanges to improve an initial tour. The program is not included in this package and has to be obtained and installed separately (see Concorde).

Additional control options: see Concorde above.

Value

An object of class TOUR.

Author(s)

Michael Hahsler

References

David Appletgate, Robert Bixby, Vasek Chvatal, William Cook (2001): TSP cuts which do not conform to the template paradigm, Computational Combinatorial Optimization, M. Junger and D. Naddef (editors), Springer.

D. Applegate, W. Cook and A. Rohe (2003): Chained Lin-Kernighan for Large Traveling Salesman Problems. INFORMS Journal on Computing, 15(1):82–92.

G.A. Croes (1958): A method for solving traveling-salesman problems. Operations Research, 6(6):791–812.

S. Lin and B. Kernighan (1973): An effective heuristic algorithm for the traveling-salesman problem. Operations Research, 21(2): 498–516.

D.J. Rosenkrantz, R. E. Stearns, and Philip M. Lewis II (1977): An analysis of several heuristics for the traveling salesman problem. SIAM Journal on Computing, 6(3):563–581.

See Also

TOUR, TSP, ATSP, write_TSPLIB, Concorde.

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
## solve a simple Euclidean TSP (using the default method)
etsp <- ETSP(data.frame(x = runif(20), y = runif(20)))
tour <- solve_TSP(etsp)
tour
tour_length(tour)
plot(etsp, tour)


## compare methods
data("USCA50")
USCA50
methods <- c("identity", "random", "nearest_insertion",
  "cheapest_insertion", "farthest_insertion", "arbitrary_insertion",
  "nn", "repetitive_nn", "two_opt")

## calculate tours
tours <- lapply(methods, FUN = function(m) solve_TSP(USCA50, method = m))
names(tours) <- methods

## use the external solver which has to be installed separately
## Not run: 
tours$concorde  <- solve_TSP(USCA50, method = "concorde")
tours$linkern  <- solve_TSP(USCA50, method = "linkern")

## End(Not run)

## register a parallel backend to perform repetitions in parallel
## Not run: 
library(doParallel)
registerDoParallel()

## End(Not run)

## add some tours using repetition and two_opt refinements
tours$'nn+two_opt' <- solve_TSP(USCA50, method="nn", two_opt=TRUE)
tours$'nn+rep_10' <- solve_TSP(USCA50, method="nn", rep=10)
tours$'nn+two_opt+rep_10' <- solve_TSP(USCA50, method="nn", two_opt=TRUE, rep=10)
tours$'arbitrary_insertion+two_opt' <- solve_TSP(USCA50)

## show first tour
tours[[1]]

## compare tour lengths
opt <- 14497 # obtained by Concorde
tour_lengths <- c(sort(sapply(tours, tour_length), decreasing = TRUE),
  optimal = opt)
dotchart(tour_lengths/opt*100-100, xlab = "percent excess over optimum")

TSP documentation built on May 23, 2019, 1:03 a.m.