Fixing Geographic Travelling Salesman Issues utilizing Python | by Mike Jones | Jul, 2023

Utilizing pyconcorde to seek out optimum options to real-world routing issues

An optimum automotive driving route between 79 UK cities. Picture by writer. Map information from OpenStreetMap.

The well-known Travelling Salesman Problem (TSP) is about discovering an optimum route between a set of nodes (cities) and returning to the place you began. It sounds easy, however is inconceivable to resolve by brute drive for giant numbers of nodes, for the reason that variety of attainable orderings of n cities is n!. Because of this for even simply 30 cities, the variety of journeys you would need to verify is 265,252,859,812,191,058,636,308,480,000,000. Massive TSP issues are impractical to resolve by brute drive, even by highly effective computer systems.

Randomly generated TSP with 10 nodes: 3,628,800 attainable routes to verify. Picture by writer.

Luckily, some algorithms have been developed that dramatically scale back the quantity of compute wanted to resolve giant TSPs. One such piece of software program, Concorde, was developed a few many years in the past to be used within the tutorial group. Though it’s fairly technical to make use of as stand-alone software program, and is meant for specialists solely, pyconcorde has been developed as a Python wrapper for Concorde. A proof of the algorithms utilized in Concorde is outdoors the scope of this text. Nevertheless, we’ll delve into the code wanted to breed these issues and their options in Python.

How would somebody go about fixing a real-world, geographical travelling salesman downside? Actual-world factors should not linked by easy 2D traces like within the picture above. As an alternative, geographical options are linked by numerous attainable routes, and people routes will change relying on whether or not somebody is strolling, biking or driving.

Why may a knowledge scientist or software program engineer wish to resolve a real-world TSP? Listed below are a couple of examples of use circumstances:

  1. An organization using couriers wants a means of calculating optimum routes via a metropolis, minimizing the time spent on the highway for every of its drivers.
  2. A tour operator wants to seek out the shortest route connecting a set of locations, inside a constrained period of time.
  3. A waste disposal firm or native authority must allocate its sources to make sure pickups are ordered in as environment friendly a fashion as attainable.

With the intention to resolve a real-world TSP, the routingpy library can be utilized to seek out routes, distances (in metres) and durations (in seconds) between geographical factors in [longitude, latitude] pairs. On this article we’ll describe the tactic that can be utilized for such an issue.

A information to fixing a geographic TSP utilizing Python is printed right here. The broad construction of the problem-solving course of is as follows:

  1. Receive a listing of n coordinates as [longitude, latitude] pairs.
  2. Use a routing service to acquire a matrix (n x n) of real-world durations between every of those coordinates, for the suitable profile (strolling, biking, driving a automotive, driving an HGV, and so on). This matrix will likely be uneven (driving from A to B is just not the precise reverse of B to A).
  3. Rework (n x n) matrix right into a symmetric matrix (2n x 2n).
  4. Feed this matrix into the Concorde solver to seek out an optimum ordering of coordinates.
  5. Create the real-world route utilizing the routing service.
  6. Visualize the outcomes on a map.
  7. Optionally, create a GPX file of the ultimate route.

Every of those steps will likely be lined intimately.

Step 1: Acquiring coordinates

For our instance, we’ll take into account the issue of driving a automotive between 79 cities within the UK. Proven beneath is a map of the UK cities in blue. An information scientist can discover coordinates in a lot of methods. If required, they are often discovered manually utilizing Google Maps or Google Earth.

79 cities within the UK. Picture by writer. Map information from OpenStreetMap.

The code construction and information used on this instance can also be out there in this GitHub repository.

Here’s a CSV file containing the coordinates of the cities (gb_cities.csv within the repo), and beneath it the code required to import it utilizing pandas.

Place Title,Latitude,Longitude
import pandas as pd
df = pd.read_csv('gb_cities.csv')
coordinates = df[['Longitude', 'Latitude']].values
names = df['Place Name'].values

Step 2: Utilizing a routing service to acquire period matrix

There are a number of routing providers out there via the routingpy library. The API from Graphhopper features a free tier which permits rate-limited use. Different routers which might be out there via routingpy are listed within the documentation.

import routingpy as rp
import numpy as np

api_key = # get a free key at
api = rp.Graphhopper(api_key=api_key)
matrix = api.matrix(areas=coordinates, profile='automotive')
durations = np.matrix(matrix.durations)

Right here is durations, a 79 x 79 matrix of the driving time in seconds between coordinates:

matrix([[    0, 10902, 30375, ..., 23380, 25233, 19845],
[10901, 0, 23625, ..., 16458, 18312, 13095],
[30329, 23543, 0, ..., 8835, 9441, 12260],
[23397, 16446, 9007, ..., 0, 2789, 7924],
[25275, 18324, 9654, ..., 2857, 0, 9625],
[19857, 13071, 12340, ..., 8002, 9632, 0]])

The driving time between cities will be decided as follows:

  1. Every row and column corresponds to a metropolis: Aberdeen is the primary row and column, Ayr the second, Basildon the third, and so forth.
  2. To search out the time between Aberdeen and Ayr, have a look at the first row, 2nd column: 10,902 seconds. The reverse time (Ayr to Aberdeen) is 10,901 seconds.
  3. Basically, the time from the i-th to the j-th metropolis is on the intersection between the i-th row and j-th column.

Discover that the matrix, as anticipated, has zeros alongside the diagonal, since every level is linked to itself with zero distance or period. Additionally, the matrix is just not fairly symmetric: driving durations between cities are unlikely to be an identical in reverse instructions, because of totally different highway layouts and visitors hotspots. They’re broadly related, although, as could be anticipated.

Step 3: Reworking uneven matrix to symmetric

Earlier than utilizing this matrix to generate an optimum ordering in pyconcorde, we have to make the matrix symmetric. A technique for remodeling an uneven TSP into symmetric TSP is described by Jonker and Volgenant (1983): Transforming asymmetric into symmetric traveling salesman problems, Operations Research Letters, 2(4), 161–163. What follows is the speculation behind this transformation. If desired, this part will be skipped (scroll all the way down to the part titled Reworking the geographic uneven TSP).

Jonker/Volgenant uneven to symmetric transformation

Under is a visualization of an uneven TSP with 3 nodes, and its distance matrix.

Uneven TSP with 3 nodes. Picture by writer.
matrix([[0, 5, 2],
[7, 0, 4],
[3, 4, 0]])

Here’s a sketch of the tactic used to rework this right into a symmetric TSP:

  1. Create new ghost nodes, A’, B’ and C’. Be a part of A to A’, B to B’ and C to C’ with distance zero.
  2. Join the nodes with weights as follows:
    A to B is now represented by A’ to B; B to A is now B’ to A.
    B to C is now B’ to C; C to B is now C’ to B.
    C to A is now C’ to A; A to C is A’ to C.
  3. Set all different edge weights to be infinite, so any algorithm doesn’t try and journey between them. Since this will likely be impractical later when utilizing pyconcorde, as a substitute set all different weights to be a lot larger than the best weight now we have. On this case, we’ll set them to equal 99.
Equal symmetric TSP with (3 x 2) nodes. Picture by writer.

Right here is the ensuing distance matrix. The ordering of the nodes within the matrix is: A, B, C, A’, B’, C’.

matrix([[ 0, 99, 99,  0,  7,  3],
[99, 0, 99, 5, 0, 4],
[99, 99, 0, 2, 4, 0],
[ 0, 5, 2, 0, 99, 99],
[ 7, 0, 4, 99, 0, 99],
[ 3, 4, 0, 99, 99, 0]])

Be aware once more that the diagonal is zero, as could be anticipated, and that the matrix is now symmetric. The unique matrix is within the bottom-left nook of the brand new matrix, and its transpose is within the top-right. In the meantime, the top-left and bottom-right elements comprise very excessive weights between nodes.

A, B and C (top-left) are not linked to one another (strictly talking, they’re linked however with very excessive as a substitute of infinite weight, for sensible functions). Because of this any algorithm is not going to search to discover a path between these nodes. Likewise, A’, B’ and C’ (bottom-right) should not linked to one another. As an alternative, the directional nature of the unique uneven community is represented right here by the weights on the unique nodes A, B and C, along with their ghosts A’, B’ and C’.

There’s a one-to-one mapping between options of the unique uneven downside and the brand new, symmetric TSP:

  • A — B — C — A corresponds to A — A’ — B — B’ — C — C’ — A
  • A — C — B — A corresponds to A — A’ — C — C’ — B — B’ — A

In every case the ghost nodes A’, B’ and C’ alternate with the unique nodes A, B and C, and every unique node is adjoining to its ‘companion’ ghost node (A is adjoining to A’, and so forth).

Reworking the geographic uneven TSP

Again to our sensible instance. We are able to create a operate to rework an uneven TSP matrix right into a symmetric one:

def symmetricize(m, high_int=None):

# if high_int not offered, make it equal to 10 instances the max worth:
if high_int is None:
high_int = spherical(10*m.max())

m_bar = m.copy()
np.fill_diagonal(m_bar, 0)
u = np.matrix(np.ones(m.form) * high_int)
np.fill_diagonal(u, 0)
m_symm_top = np.concatenate((u, np.transpose(m_bar)), axis=1)
m_symm_bottom = np.concatenate((m_bar, u), axis=1)
m_symm = np.concatenate((m_symm_top, m_symm_bottom), axis=0)

return m_symm.astype(int) # Concorde requires integer weights

symmetricize(durations) returns:

matrix([[     0, 461120, 461120, ...,  23397,  25275,  19857],
[461120, 0, 461120, ..., 16446, 18324, 13071],
[461120, 461120, 0, ..., 9007, 9654, 12340],
[ 23397, 16446, 9007, ..., 0, 461120, 461120],
[ 25275, 18324, 9654, ..., 461120, 0, 461120],
[ 19857, 13071, 12340, ..., 461120, 461120, 0]])

This 158 x 158 matrix comprises a replica of durations within the backside left and a transposed copy within the high proper. The excessive worth of 461,120 (10 instances the utmost worth in durations) signifies that, for sensible functions, nodes with this period should not linked.

This matrix can lastly be fed into pyconcorde to calculate an optimum path.

Step 4: Utilizing the Concorde solver

Putting in pyconcorde

Run the next instructions to put in pyconcorde (set up is obtainable in Linux or Mac OS, however not in Home windows at current):

virtualenv venv                                  # create digital setting
supply venv/bin/activate # activate it
git clone # clone git repo
cd pyconcorde # change listing
pip set up -e . # set up pyconcorde

Fixing the TSP in Python

Now we will import from concorde in a Python script.

from concorde.downside import Downside
from concorde.concorde import Concorde

def solve_concorde(matrix):
downside = Downside.from_matrix(matrix)
solver = Concorde()
resolution = solver.resolve(downside)
print(f'Optimum tour: {resolution.tour}')
return resolution

Our symmetric durations matrix will be fed into solve_concorde().

durations_symm = symmetricize(durations)
resolution = solve_concorde(durations_symm)

Right here is the print output:

Optimum tour: [0, 79, 22, 101, 25, 104, 48, 127, 68, 147, 23, 102, 58, 137, 7, 86, 39, 118, 73, 152, 78, 157, 36, 115, 42, 121, 62, 141, 16, 95, 20, 99, 51, 130, 40, 119, 19, 98, 59, 138, 50, 129, 54, 133, 27, 106, 10, 89, 4, 83, 66, 145, 33, 112, 14, 93, 2, 81, 45, 124, 32, 111, 11, 90, 29, 108, 34, 113, 24, 103, 8, 87, 17, 96, 56, 135, 64, 143, 61, 140, 75, 154, 52, 131, 71, 150, 18, 97, 3, 82, 9, 88, 74, 153, 55, 134, 72, 151, 28, 107, 12, 91, 70, 149, 65, 144, 35, 114, 31, 110, 77, 156, 63, 142, 41, 120, 69, 148, 6, 85, 76, 155, 67, 146, 15, 94, 44, 123, 47, 126, 60, 139, 57, 136, 38, 117, 13, 92, 5, 84, 43, 122, 49, 128, 46, 125, 21, 100, 1, 80, 30, 109, 53, 132, 37, 116, 26, 105]

This resolution exhibits the ordering of nodes within the optimum tour. Be aware that this resolution, as anticipated above, comprises unique nodes (numbered 0 to 78) alternating with their companion ghost nodes (79 to 157):

  • 0 is partnered with 79,
  • 22 with 101,
  • 25 with 104, and so forth…

This means that the answer has labored accurately.

Step 5: Creating the real-world route

The following step is to select alternate components of the answer (the nodes equivalent to the unique 79 cities), then order the coordinates accordingly.

# choose alternate components: these correspond to the originals
tour = resolution.tour[::2]

# order the unique coordinates and names
coords_ordered = [coordinates[i].tolist() for i in tour]
names_ordered = [names[i] for i in tour]

Listed below are the primary few metropolis names in names_ordered, (the true ordering of the cities within the optimum tour):

'Newcastle Upon Tyne',

Now we add again within the first metropolis to make a whole looped tour, and at last receive the ultimate route utilizing the Graphhopper instructions API.

# add again within the first for an entire loop
coords_ordered_return = coords_ordered + [coords_ordered[0]]

# receive full driving instructions for the ordered loop
instructions = api.instructions(areas=coords_ordered_return, profile='automotive')

Step 6: Visualization on a map

Seeing the ultimate route on a map will allow us to be assured within the outcome, in addition to permitting us to make use of the answer in a sensible setting. The next code will show a folium map which will be saved to HTML.

import folium
def generate_map(coordinates, names, instructions):

# folium wants lat, lengthy
coordinates = [(y, x) for (x, y) in coordinates]
route_points = [(y, x) for (x, y) in directions.geometry]
lat_centre = np.imply([x for (x, y) in coordinates])
lon_centre = np.imply([y for (x, y) in coordinates])
centre = lat_centre, lon_centre

m = folium.Map(location=centre, zoom_start=1, zoom_control=False)

# plot the route line
folium.PolyLine(route_points, coloration='purple', weight=2).add_to(m)

# plot every level with a hover tooltip
for i, (level, title) in enumerate(zip(coordinates, names)):
tooltip=f'{i}: {title}',

custom_tile_layer = folium.TileLayer(
attr='CartoDB Positron',
opacity=0.7 # Modify opacity to manage the extent of greying out


sw = (np.min([x for (x, y) in coordinates]), np.min([y for (x, y) in coordinates]))
ne = (np.max([x for (x, y) in coordinates]), np.max([y for (x, y) in coordinates]))
m.fit_bounds([sw, ne])

return m

generate_map(coords_ordered, names_ordered, instructions).save('gb_cities.html')

The result’s proven on the high of this text. Click here to view as an interactive map. It’s attainable to zoom in to the map to see extra element and to hover over particular person cities which can reveal their quantity within the tour sequence. Under is a zoomed-in a part of the map exhibiting the route passing via Sheffield (between Lincoln and Chesterfield on the optimum tour).

Picture by writer. Map information from OpenStreetMap.

Step 7: Non-obligatory: Making a GPX file

If the calculated route must be adopted in real-life, for example on a tool with a GPS (reminiscent of a telephone or automotive navigation system), a GPX will be created. This isn’t a part of the optimization downside, however is an non-obligatory extra step out there if you wish to save the path to a file. The GPX file is created from the instructions variable:

def generate_gpx_file(instructions, filename):
gpx_template = """<?xml model="1.0" encoding="UTF-8"?>
<gpx model="1.1" xmlns=""

trkseg_template = """
<trkpt lat="{}" lon="{}"/>

trkseg_elements = ""
for level in instructions.geometry:
trkseg_elements += trkseg_template.format(level[1], level[0])

gpx_data = gpx_template.format(trkseg_elements)

with open(filename, 'w') as file:

generate_gpx_file(instructions, 'gb_cities.gpx')

The GPX file for this downside will be downloaded here.

We’ve seen how we will mix the next components to resolve real-world geographic travelling salesman issues:

  1. Instructions and period matrices from the routingpy library, specifying an applicable profile (mode of transport).
  2. Environment friendly and highly effective Concorde solver through the pyconcorde wrapper, to supply an optimum route.
  3. Visualization utilizing folium to create a map.

The driving tour proven above is a convincing resolution to the 79-city travelling salesman downside, and in response to the Concorde solver is provably ‘optimum’. Since we’re working with real-world information, nonetheless, the tip result’s solely pretty much as good because the enter. We’re relying that the point-to-point durations matrix obtained from routingpy is consultant of the real-world. In actuality, the time taken to stroll, cycle or drive between factors will rely on the time of day, or day of the week. This can be a limitation of the tactic that we’ve used. A method of being extra assured in the long run outcome could be to strive the identical methodology with an alternative routing service. Every routing service (Graphhopper, ORS, Valhalla, and so forth) has its personal API which could possibly be used for a TSP downside such because the one described right here, and the outcomes could possibly be in contrast from totally different providers.

Regardless of the real-world limitations of fixing an issue reminiscent of this, the methodology above offers a great start line for a salesman or courier needing to make their means spherical a metropolis in as environment friendly a fashion as attainable or a vacationer hoping to catch as many sights as attainable on their tour. By visualizing the outcomes on an interactive map and storing the route as a GPX file, the answer is beneficial by the tip person, not simply the information scientist who applied the code.

8 Strategies to Mannequin Seasonality | by Vitor Cerqueira | Jul, 2023

Araucana XAI: Why Did AI Get This One Incorrect? | by Tommaso Buonocore