some changes

Author: bornabr

README.md

@@ -0,0 +1,59 @@
+# TSP Problem with Genetic Algorithm
+## Data
+The input of this problem files with `tsp` format. I used the `tsplib95` library for weighted test cases (`bayg29`) and a self-written loader for test cases with coordinations.
+
+## Algorithm
+At first, we generate our initial population with random routes(permutations).
+
+The main process is a loop with several sections. At the end of every iteration in this loop, a new generation is born, preferably with better individuals.
+
+Here is each section of this process:
+
+### Selection
+The first thing we do with the previous generation is to select a subset of them as our parents. The primary consideration here is to make a balance between elites and vulgar to prevent overfitting or underfitting.
+
+The algorithm used for this matter is tournament selection. For this, we choose a subset of the population by random, then we sort them based on their fitness and select a number of them as parents.
+
+### Breeding
+With the parents being chosen, it is time for mating them with each other. Two individuals create a child with ordered recombination.
+
+For choosing each pair, we shuffle parents and mate them by random. We repeat this shuffling until the number of our children has reached the population size.
+
+### Mutation
+We have a parameter called the `mutation_rate`, which specifies the probability of mutation.
+
+The mutations in this problem are done by swapping two cities(genes) in a permutation.
+
+### Replacement
+The final step in the loop is to specify the next generation. We have considered a mix of children and parents because we want to keep the best parents for the next generation.
+We replace a fixed number of worst children with the best parents.
+
+## Test
+### bayg29
+My best distances were between 1600 to 1800 and here is one of them:
+```
+Epoch 0 :       Population total fitness: 0.021402390435083837  Best fitness: 0.000275178866263071      Least Distance: 3634
+Epoch 1 :       Population total fitness: 0.022400741669908814  Best fitness: 0.00028694404591104734    Least Distance: 3485
+Epoch 2 :       Population total fitness: 0.02321754970711249   Best fitness: 0.000299311583358276      Least Distance: 3341
+Epoch 3 :       Population total fitness: 0.024596784139334586  Best fitness: 0.00030321406913280777    Least Distance: 3298
+Epoch 4 :       Population total fitness: 0.025508500567304724  Best fitness: 0.00031655587211142766    Least Distance: 3159
+Epoch 5 :       Population total fitness: 0.02567080676446863   Best fitness: 0.00031655587211142766    Least Distance: 3159
+Epoch 6 :       Population total fitness: 0.026095669980930384  Best fitness: 0.00031826861871419476    Least Distance: 3142
+Epoch 7 :       Population total fitness: 0.026213325051129905  Best fitness: 0.0003176620076238882     Least Distance: 3148
+...
+Epoch 92 :      Population total fitness: 0.046347892289803405  Best fitness: 0.0005633802816901409     Least Distance: 1775
+Epoch 93 :      Population total fitness: 0.04640791685328948   Best fitness: 0.0005633802816901409     Least Distance: 1775
+Epoch 94 :      Population total fitness: 0.0463885029225474    Best fitness: 0.000572737686139748      Least Distance: 1746
+Epoch 95 :      Population total fitness: 0.04655714750065836   Best fitness: 0.000572737686139748      Least Distance: 1746
+Epoch 96 :      Population total fitness: 0.04607831403892826   Best fitness: 0.000572737686139748      Least Distance: 1746
+Epoch 97 :      Population total fitness: 0.04588737374610269   Best fitness: 0.000572737686139748      Least Distance: 1746
+Epoch 98 :      Population total fitness: 0.04635046560907124   Best fitness: 0.000572737686139748      Least Distance: 1746
+Epoch 99 :      Population total fitness: 0.0466206756566057    Best fitness: 0.000572737686139748      Least Distance: 1746
+Best Answer:
+[13, 4, 20, 10, 2, 29, 3, 26, 6, 12, 9, 5, 21, 28, 1, 24, 8, 23, 27, 16, 7, 25, 19, 15, 18, 17, 14, 11, 22] 1746
+```
+
+
+## Considerations
+- Because we want to optimize our answers based on the total distance of the cycle, our fitness function is `1/total_distance` (so we try to maximize fitness).
+

__pycache__/loader.cpython-38.pyc

@@ -0,0 +1,34 @@
+class Problem:
+	def __init__(self, filepath):
+		f = open(filepath, 'r')
+		lines = f.readlines()
+		self.coordinations = []
+		start = False
+		for line in lines:
+			if start:
+				line = line.strip().split()
+				if len(line) < 3:
+					break
+				self.coordinations.append((float(line[1]), float(line[2])))
+				continue
+			
+			if 'DIMENSION' in line:
+				line = line.split(':')
+				self.dimension = int(line[1].strip())
+				continue
+			
+			if 'NODE_COORD_SECTION' in line:
+				start = True
+		self.calculate_weights()
+	
+	def calculate_weights(self):
+		self.weights = []
+		for first_coordination in self.coordinations:
+			weights = []
+			for second_coordination in self.coordinations:
+				weights.append((((first_coordination[0]-second_coordination[0])**2) +
+				        ((first_coordination[1]-second_coordination[1])**2))**0.5)
+			self.weights.append(weights)
+	
+	def get_weight(self, start, end):
+		return self.weights[start-1][end-1]

__pycache__/utils.cpython-38.pyc

@@ -1,31 +1,14 @@
-import tsplib95
-
-from utils import initial_population
-from utils import generate_genration
-
-
-def TSP_GA(filename, n_generations, population_size, tournament_size, parents_size, mutation_rate, elite_size):
-	problem = tsplib95.load('./problems/' + filename)
-	
-	population = initial_population(problem, population_size)
-	
-	for i in range(n_generations):
-		population = generate_genration(
-			i, population, tournament_size, parents_size, mutation_rate, elite_size)
-	
-	print([x.distance for x in population])
-	
-	print('Best Answer:')
-	print(population[0].permutation, population[0].distance)
+from loader import Problem
+from utils import TSP_GA
 
 def bayg29():
 	population_size = 100
 
 	tournament_size = 90
 
-	parents_size = 40
+	parents_size = 60
 
-	elite_size = 20
+	elite_size = 30
 
 	mutation_rate = 0.01
 
@@ -35,55 +18,42 @@ def bayg29():
 	       parents_size, mutation_rate, elite_size)
 
 
-def bayg29():
-	population_size = 100
+def gr229():
+	population_size = 200
 
-	tournament_size = 90
+	tournament_size = 160
 
-	parents_size = 40
+	parents_size = 90
 
-	elite_size = 20
+	elite_size = 40
 
 	mutation_rate = 0.01
 
-	n_generations = 100
+	n_generations = 10000
 
-	TSP_GA('bayg29.tsp', n_generations, population_size, tournament_size,
-	       parents_size, mutation_rate, elite_size)
-
-
-def gr229():
-	population_size = 100
-
-	tournament_size = 90
-
-	parents_size = 70
-
-	elite_size = 30
-
-	mutation_rate = 0.05
-
-	n_generations = 100
+	problem = Problem('./problems/gr229.tsp')
 
 	TSP_GA('gr229.tsp', n_generations, population_size, tournament_size,
-	       parents_size, mutation_rate, elite_size)
+	       parents_size, mutation_rate, elite_size, problem)
 
 
 def pr1002():
-	population_size = 100
+	population_size = 200
 
-	tournament_size = 90
+	tournament_size = 140
 
-	parents_size = 50
+	parents_size = 100
 
-	elite_size = 35
+	elite_size = 80
 
-	mutation_rate = 0.05
+	mutation_rate = 0.1
 
 	n_generations = 150
 
+	problem = Problem('./problems/pr1002.tsp')
+
 	TSP_GA('pr1002.tsp', n_generations, population_size, tournament_size,
-	       parents_size, mutation_rate, elite_size)
+	       parents_size, mutation_rate, elite_size, problem)
 
 # bayg29()
 

loader.py

@@ -1,4 +1,5 @@
 import random
+import tsplib95
 
 from cycle import Cycle
 
@@ -110,3 +111,17 @@ def generate_genration(epoch, previous_population, tournament_size, parents_size
 	      eval_[0], "\tBest fitness:", eval_[1], "\tLeast Distance:", eval_[2])
 
 	return next_population
+
+
+def TSP_GA(filename, n_generations, population_size, tournament_size, parents_size, mutation_rate, elite_size, problem=None):
+	if problem is None:
+		problem = tsplib95.load('./problems/' + filename)
+
+	population = initial_population(problem, population_size)
+
+	for i in range(n_generations):
+		population = generate_genration(
+			i, population, tournament_size, parents_size, mutation_rate, elite_size)
+
+	print('Best Answer:')
+	print(population[0].permutation, population[0].distance)