Path: blob/master/ga/tsp_solver/__pycache__/tspga.cpython-35.pyc
2594 views
5O@Y� � @ sm d d l Z d d l Z d d l Z d d l j Z d d l m Z d d l
m Z Gd d � d e � Z
d S)� N)�
namedtuple)� combinationsc @ s� e Z d Z d Z d d � Z d d � Z d d � Z d d � Z d
d � Z d d
� Z d d � Z
d d � Z d d � Z d S)�TSPGAa
Travel Salesman Problem using Genetic Algorithm
Parameters
----------
generation : int
number of iteration to train the algorithm
population_size : int
number of tours in the population
retain_rate : float between 0 ~ 1
the fraction of the best tour (shortest total distance)
in the population to retain, which is then used in the
crossover and mutation staget to generate the children
for the next generation
mutate_rate : float between 0 ~ 1
the probability that each tour will mutate
Example
-------
%matplotlib inline
import pandas as pd
from tsp_solver import TSPGA
import matplotlib.pyplot as plt
# toy dataset
tsp_data = pd.read_table(
'TSP_berlin52.txt',
skiprows = 1, # the first row is simply the number of cities
header = None,
names = [ 'city', 'x', 'y' ],
sep = ' '
)
# specify the parameters and fit the data
tsp_ga = TSPGA(
generation = 5000,
population_size = 250,
retain_rate = 0.5,
mutate_rate = 0.25
)
tsp_ga.fit(tsp_data)
# distance convergence plot, and the best tour's distance
# and the corresponding city tour
tsp_ga.convergence_plot()
tsp_ga.best_tour
Reference
---------
http://www.theprojectspot.com/tutorial-post/applying-a-genetic-algorithm-to-the-travelling-salesman-problem/5
c C s; | | _ | | _ | | _ | | _ t | | � | _ d S)N)�
generation� retain_rate� mutate_rate�population_size�int�
retain_len)�selfr r r r � r �6/Users/r631854/machine-learning/ga/tsp_solver/tspga.py�__init__? s
zTSPGA.__init__c C s� | j d | _ | d d g j | _ d d � t | d � D� | _ | j � | _ t d d d g � | _ | j
d | d � } g | _ x< t | j
� D]+ } | j | � \ } } | j j | � q� W| j | j
d
| _ d | _ | S) a
fit the genetic algorithm on the tsp data
Parameters
----------
tsp_data : DataFrame
contains the 'city' and its 'x', 'y' coordinate, note that
the column name must match or the code will break (ordering
of the column does not matter)
r �x�yc S s i | ] \ } } | | � q Sr r )�.0�index�cityr r r
�
<dictcomp>Y s z TSPGA.fit.<locals>.<dictcomp>r � tour_info�dist�tour� T)�shape�city_num�values�cities� enumerate� city_dict�_compute_pairwise_distance�pairwise_distancer r �_generate_tours�generation_history�ranger �_evolve�append� best_tour� is_fitted)r �tsp_data�
population�_�generation_bestr r r
�fitG s
z TSPGA.fitc C s� t j | j | j f � } xd t | j j � d � D]J \ } } t j j | j | | j | � } | | | | <| | | <q4 W| S)z{
readable but not optimized way of computing and storing
the symmetric pairwise distance for between all city pairs
� ) �np�zerosr r r r �linalg�normr )r r �i1�i2�distancer r r
r n s
%$ z TSPGA._compute_pairwise_distancec s� g } xC t � j � D]2 } | j j � } t j j | � | j | � q W� f d d � | D� } � f d d � t | | � D� } t | � } | S)z�
or in genetic algorithm terms, generate the populations.
generate a new random tour with the size of the population_size
and compute the distance for each tour
c s" g | ] } � j d | � � q S)r )�_compute_tour_distance)r �t)r r r
�
<listcomp>� s z)TSPGA._generate_tours.<locals>.<listcomp>c s( g | ] \ } } � j | | � � q Sr )r )r r r )r r r
r7 � s )
r# r r �copyr. �random�shuffler% �zip�sorted)r r Ztoursr* r Z
tours_distr) r )r r
r! { s zTSPGA._generate_toursc C s� | j | d } | j d } | j | | } | j | | } xd t | � D]V \ } } | j | } | d } | j | | }
| | j | |
7} | | k rM PqM W| S)z�
1. compute the total distance for each tour,
note that tour stores the name of the city, thus you need to map it
with the city_dict to access the pairwise distance.
2. initialize the distance with the last city to the first city's distance
r r )r r r r ) r r Z
first_city�
last_indexZ last_cityZ tour_distr Z
city_indexr Z
next_indexZ next_cityr r r
r5 � s
z TSPGA._compute_tour_distancec s | j � } | d � j � } d d � | D� } g } xD t | � � j k r~ � j | � } � j | � } | j | � q; W� f d d � | D� } � f d d � t | | � D� } y0 | j | � t | � } | d � j � } Wn t
k
r| } Yn X| d } | | f S)zt
core method that does the crossover, mutation to generate
the possibly best children for the next generation
Nc S s g | ] } | j � q Sr )r )r �pr r r
r7 � s z!TSPGA._evolve.<locals>.<listcomp>c s" g | ] } � j d | � � q S)r )r5 )r �c)r r r
r7 � s c s( g | ] \ } } � j | | � � q Sr )r )r r r )r r r
r7 � s r ) r8 r
�lenr �
_crossover�_mutater% r; �extendr< �
ValueError) r r) Zpopulation_backup�parent�children�childZ
children_distZpopulation_childrenr+ r )r r
r$ � s&
z
TSPGA._evolvec C s� t j t | j � d d �\ } } | | | | } } t j t | j � d d �\ } } | | k rv | | } } t | | � } t j | | | d d �} | | }
t j |
d | � | | |
| d � f } | S)z(randomly select two parent and mate them�kr- �invertTN) r9 �sampler# r
r �slicer. �in1d�r_) r rE �index1�index2�male�femaleZ pos_startZpos_end�subset�booleanZ not_in_malerG r r r
rA � s $$
.zTSPGA._crossoverc C sZ | j t j � k rV t j t | j � d d �\ } } | | | | | | <| | <| S)z[
randomly swap the position of two cities if
the the mutuate_rate's threshold is met
rH r- )r r9 rJ r# r )r rG Zswap1Zswap2r r r
rB � s $ z
TSPGA._mutatec C s� | j s t d � d d � | j D� } t j t d t | � d � | d � t j d � t j d � t j d � t j
� t j � d S)
zX
convergence plot showing the decrease of each generation's
best tour's distance
z,you have not fitted the algorithm using .fitc S s g | ] } | j � q Sr )r )r �gr r r
r7 � s z*TSPGA.convergence_plot.<locals>.<listcomp>r �-zDistance Convergence PlotZ IterationZDistanceN) r' rD r" �plt�plotr# r@ �title�xlabel�ylabel� tight_layout�show)r r r r r
�convergence_plot� s
&
zTSPGA.convergence_plotN)
�__name__�
__module__� __qualname__�__doc__r r, r r! r5 r$ rA rB r] r r r r
r s 6 '
+ r )r9 �numpyr. �pandas�pd�matplotlib.pyplot�pyplotrV � collectionsr � itertoolsr �objectr r r r r
�<module> s