pip install tsp-solutions

The travelling salesman problem is an np-hard problem with application in supply chain and computer science. The module has two parts:

tsp_exact uses PuLP module to formulate the problem and CPLEX, GUROBI, COIN_CMD, and PULP_solver to find the exact solution of the TSP. To setup an external solver follow this link.

heuristics & metaheuristic functions described later are local search algorithms which can be used for large instances.

Dependencies include pulp, numpy, pandas, scipy, and copy.

If you find an error please raise an issue and i will respond.

1. dynamic programming algorithm

1. matrix: is a NxN cost matrix between the points that have to be visited by the nodes with 2 requirements:

a. The row and column index of this matrix should be INTEGER

b. Depot is indexed by 0, i.e. Row 0 represents the Depot.

    0     1   2     3   4
0   0  21.0  15  21.0  10
1  21   0.0   5   0.5  11
2  15   5.0   0   5.0   5
3  21   0.5   5   0.0  10
4  10  11.0   5  10.0   0

2. method: The type of solver to use. Default is "PULP_CBC_CMD". Other solvers available are: "CPLEX_CMD", "GUROBI_CMD", and "COIN_CMD".
3. message: PuLP will display calculation summary if message is set to 1. Default value is 0 (suppress summary).
4. route: Initial route estimate (only for improvement heuristics & metaheuristics)
5. trials: Number of local search to be made, default is 5000 (only for improvement heuristics & metaheuristics)

1. route[List]: The shortest route coving all the nodes. (The route starts from the depot and ends at the depot)
2. route_sum[Scalar]: Total cost incurred by the route.

For small instances with $\leq15$ nodes, the tsp_exact function will give the exact solution. For instances $> 15$ nodes, heuristic solutions are preferred due to high computation time. (Remember, tsp is a NP-hard problem with O(n!))

We define the following heuristics for tsp:

1. Nearest neighbor heuristics: The nearest node to the tour is added to the tour. See Greedy Search
2. Nearest insertion heuristics: The nearest node to the tour is inserted in the tour at an arc with minimum cost
3. Cheapest insertion heuristics: Modified version of Nearest insertion heuristics which checks all nodes for insertion instead of the nearest node
4. Farthest insertion heuristics: The farthest node to the tour is inserted in the tour at an arc with minimum cost
5. MST heuristics: Minimum spanning tree is created for the network and repeated nodes are removed to form the route See Minimum Spanning Tree

6. 2-Opt exchange: A local search algorithm which exchanges 2 route nodes and stores the solution if it has lower objective value. See 2-Opt

7. Tabu search: See Tabu Seach This is a simple version of tabu search with 2-opt exchange used for local search. A more comprehensive version will be released soon.

Below is the mathematical formulation for the exact solution of tsp which is executed by tsp_exact() function

$\mathbb{N}$ is the set of all customer node subset $i$ and $j$

We will use binary variable $x_{ij}$

$x_{ij}$ will take the value 1 if truck travels from node $i$ to node $j$, 0 otherwise. $i\in\mathbb{N}$ and $j\in\mathbb{N}$

Other variables are:

$u_{i}$ will take the value of the order of node $i$ in the final route of truck. $i\in\mathbb{N}^{}$

$t_{i}$ represents the arrival time for truck at node $i$.

$tt_{ij}$ represents the truck travel time between nodes $i$ and $j$.

$M$ is a very large number

Objective: Minimize the total time to visit all nodes

$$ Obj=min{\sum_{i}t_{i}} $$

Constraint 1: All nodes have to be visited by truck exactly once

$$ \sum_{i}x_{ij}=1\quad j\in\mathbb{N}$$

Constraint 2: Truck leaves depot D and comes back to depot D' exactly once $i=D,D'$

$$ \sum_{j}x_{ij}=1 $$

$$ \sum_{j}x_{ji}=1 $$

Constraint 3: If truck arrives at node j then it should also leave node j.

$$ \sum_{i}x_{ij}=\sum_{k}x_{jk} \quad j\in\mathbb{N}$$

Constraint 4: Avoiding sub-tours for truck

$$ u_{j}-u_{k}-1\leq M(1-x_{jk}) \quad j,k\in\mathbb{N}$$

Constraint 5: We will add travel time $tt_{ij}$ to arrival time at node $i$ to get arrival time at node $j$ if truck travels in $ij$ path

$$ t_{j}\geq t_{i}+tt_{ij}-M(1-x_{ij}) \quad i,j\in\mathbb{N}$$

import pandas as pd
import tsp_solutions  # import tsp-solutions module

matrix = pd.read_csv('Hamilton_road_distance.csv')
tsp = tsp_solutions.tsp() # calling class tsp
matrix=
        0       1       2       3       4       5       6       7       8
0   0.000  14.015  14.411  14.749  21.253  19.333   9.370  10.186  15.054
1  14.003   0.000   5.317  30.438   9.109  16.033   6.130   5.209   5.062
2  14.022   5.251   0.000  30.411   6.656  14.415   8.705   4.708   1.047
3  14.755  30.741  30.746   0.000  37.979  21.296  26.096  26.521  31.389
4  21.384   8.819   5.854  37.818   0.000  21.869  14.803  25.428   6.925
5  18.620  16.179  15.038  21.287  21.128   0.000  23.369  10.812  15.680
6   9.890   5.853   8.274  26.325  14.396  23.082   0.000  13.935   8.031
7   9.987   5.599   4.458  26.377  10.709  10.938   7.619   0.000   5.100
8  15.122   6.061   1.372  31.512   7.284  15.516   8.907   5.354   0.000

result = tsp.tsp_exact(matrix, method = 'CPLEX_CMD')
print(result)

([0, 6, 1, 4, 2, 8, 7, 5, 3, 0], 83.5669999999999)

initial_route = [0,1,2,3,4,5,6,7,8,0]
result = tsp.tabu_search(initial_route,matrix, trials = 15000)
print(result)

([0, 6, 1, 4, 2, 8, 7, 5, 3, 0], 83.56700000000001)

initial_route = [0,1,2,3,4,5,6,7,8,0]
result = tsp.Opt2(initial_route,matrix, trials = 15000)
print(result)

([0, 6, 5, 2, 7, 3, 4, 8, 1, 0], 84.31700000000001)

result = tsp.MST_heuristic(matrix)
print(result)

([0, 3, 6, 1, 8, 2, 4, 7, 5, 0], 114.774)

result = tsp.fi_heuristic(matrix) # farthest insertion
print(result)

([0, 3, 6, 1, 4, 2, 8, 7, 5, 0], 97.62)