genetic-algorithm/gen_algorithm.py at master · orzechdev/genetic-algorithm · GitHub

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
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
import numpy as np
import matplotlib.pyplot as plt
from utils import print_title, plot_polynomial, random_one, unpack_bits, mutate_bits


def start(negative_points, positive_points, y_polynomial, initial_pop, param_crossover_probability, param_mutation_probability, param_generation_number):
    pop_size = initial_pop.shape[1]

    print_title('GA: start - polynomial degree 2')
    print_title('GA: initialise population of functions arguments')
    print(initial_pop)
    print_title('GA: evaluate negative and positive points groups')
    args_number = initial_pop.shape[0]
    args_pop_selected = np.zeros((args_number, 0), dtype=np.float16)
    args_pop_all, args_pop_fitness_values = evaluate(initial_pop, negative_points, positive_points, y_polynomial, args_pop_selected)
    print(args_pop_fitness_values)
    print_title('GA: selection, sort population by fitness values')
    args_pop_selected, best_value, worst_value, mean_value = selection_sort(args_pop_all, args_pop_fitness_values, pop_size)

    resulting_fitness_values = np.zeros((1, param_generation_number), dtype=np.float16)

    print(args_pop_selected)
    resulting_fitness_values[0] = best_value
    print('%.2f\t%.2f\t%.2f' % (best_value.astype(float), mean_value.astype(float), worst_value.astype(float)), end='\n')

    for i in range(param_generation_number - 1):
        # print(i)
        args_pop_crossovered = crossover(args_pop_selected, param_crossover_probability)
        args_pop_mutated = mutation(args_pop_crossovered, param_mutation_probability)
        args_pop_all, args_pop_fitness_values = evaluate(args_pop_mutated, negative_points, positive_points, y_polynomial, args_pop_selected)
        args_pop_selected, best_value, worst_value, mean_value = selection_sort(args_pop_all, args_pop_fitness_values, pop_size)
        resulting_fitness_values[0, i+1] = best_value
        print('%.2f\t%.2f\t%.2f' % (best_value.astype(float), mean_value.astype(float), worst_value.astype(float)), end='\n')

    args = args_pop_selected[:, 0][::-1]
    x = np.linspace(-10, 20, 1000)
    y = y_polynomial(x, *args)
    np.set_printoptions(suppress=True)
    plt.plot(x, y, label=str(np.around(args, decimals=3)[::-1]).strip('[]'))
    # plt.legend()
    plt.title('Polynomial of ' + str(len(args) - 1) + ' degree', loc='left')
    plt.title(str(best_value) + '% fit', loc='right')

    print('\n')
    print(args_pop_selected)

    return resulting_fitness_values


def evaluate(args_pop, negative_points, positive_points, y_polynomial, args_pop_selected):

    args_pop_fitness_values = np.empty((0, 2), dtype=np.float16)

    args_pop_all = np.insert(args_pop, [1], args_pop_selected, axis=1)

    # print(args_pop)
    # print(args_pop_selected)

    for n in range(0, args_pop_all.shape[1]):
        args = args_pop_all[:, n][::-1]
        # plot_polynomial(a, b, c)
        fitness_value = 0
        for i in range(0, negative_points.shape[1]):
            y_value = y_polynomial(negative_points[0, i], *args)
            # y_polynomial = (a * (negative_points[0, i] ** 2)) + (b * (negative_points[0, i])) + c
            if y_value > negative_points[1, i]:
                fitness_value += 1
        for i in range(0, positive_points.shape[1]):
            y_value = y_polynomial(positive_points[0, i], *args)
            # y_polynomial = (a * (positive_points[0, i] ** 2)) + (b * (positive_points[0, i])) + c
            if y_value < positive_points[1, i]:
                fitness_value += 1
        args_pop_fitness_values = np.append(args_pop_fitness_values, fitness_value)

    # args_pop_evaluated = np.concatenate((args_pop_negative, args_pop_positive), axis=1)
    return args_pop_all, args_pop_fitness_values


def selection_sort(args_pop, args_pop_fitness_values, pop_size_selected):
    args_pop_fitness_values_selected = np.empty((0, 2), dtype=np.float16)

    # print(args_pop)

    args_number = args_pop.shape[0]

    args_pop_selected = np.zeros((args_number, pop_size_selected), dtype=np.float16)
    args_pop_fitness_values_sort_sequence = np.argsort(args_pop_fitness_values)[::-1][:args_pop.shape[1]]
    best_value = args_pop_fitness_values[args_pop_fitness_values_sort_sequence[0]]
    worst_value = args_pop_fitness_values[args_pop_fitness_values_sort_sequence[len(args_pop_fitness_values_sort_sequence) - 1]]
    mean_value = np.mean(args_pop_fitness_values)

    for i in range(0, pop_size_selected):
        selected_i = args_pop_fitness_values_sort_sequence[i]
        for j in range(0, args_number):
            args_pop_selected[j, i] = args_pop[j, selected_i]
        args_pop_fitness_values_selected = np.append(args_pop_fitness_values_selected, args_pop_fitness_values[selected_i])

    # print(args_pop_fitness_values_selected)
    return args_pop_selected, best_value*100/120, worst_value*100/120, mean_value*100/120


def crossover(args_pop_selected, crossover_probability):
    args_number = args_pop_selected.shape[0]
    args_pop_crossovered = np.zeros((args_number, 0), dtype=np.float16)
    # crossover_point = np.uint8(1/2)
    # args_pop_selected_bits = np.empty((3, args_pop.shape[1], 8), dtype=np.uint8)
    # args_pop_selected_bits[0] = unpack_bits(args_pop_selected[0], 8)
    # args_pop_selected_bits[1] = unpack_bits(args_pop_selected[1], 8)
    # args_pop_selected_bits[2] = unpack_bits(args_pop_selected[2], 8)
    # print(args_pop_selected_bits)

    # individual_current = 0

    for i in range(0, args_pop_selected.shape[1]):
        for j in range(0, args_pop_selected.shape[1]):
            if random_one(crossover_probability):
                args = np.zeros((args_number, 1), dtype=np.float16)
                for k in range(0, args_number):
                    if k < args_number - 2:
                        a_i_bits = np.binary_repr(int(args_pop_selected[k, i]*1000))
                        a_j_bits = np.binary_repr(int(args_pop_selected[k, j]*1000))
                        # print(a_i_bits + ' - ' + a_j_bits)
                        a_i_bits_len = len(a_i_bits)
                        a_i_bits_half = a_i_bits if a_i_bits_len == 1 else a_i_bits[:int(a_i_bits_len/2)]
                        a_j_bits_len = len(a_j_bits)
                        a_j_bits_half = a_j_bits if a_j_bits_len == 1 else a_j_bits[int(a_j_bits_len - (a_j_bits_len/2)):]
                        # print(a_i_bits_half + ' + ' + a_j_bits_half)
                        bits_crossed = a_i_bits_half + a_j_bits_half
                        # args_pop_crossovered[k, individual_current] = int(bits_crossed, 2)/100
                        args[k, 0] = int(bits_crossed, 2)/1000
                        # print(str(i) + ',' + str(j) + ' -> ' + bits_crossed + ' -> ' + args_pop_selected[k, i].astype(str) + ' + ' + args_pop_selected[k, j].astype(str) + ' -> ' + args[k, 0].astype(str))
                    else:
                        p_i_bits = np.binary_repr(int(args_pop_selected[k, i]))
                        p_j_bits = np.binary_repr(int(args_pop_selected[k, j]))
                        p_i_bits_len = len(p_i_bits)
                        p_i_bits_half = p_i_bits if p_i_bits_len == 1 else p_i_bits[:int(p_i_bits_len/2)]
                        p_j_bits_len = len(p_j_bits)
                        p_j_bits_half = p_j_bits if p_j_bits_len == 1 else p_j_bits[int(p_j_bits_len - (p_j_bits_len/2)):]
                        # print(p_i_bits_half + ' - ' + p_j_bits_half)
                        bits_crossed = p_i_bits_half + p_j_bits_half
                        # args_pop_crossovered[k, individual_current] = int(bits_crossed, 2)
                        args[k, 0] = int(bits_crossed, 2)
                    # if args_pop_crossovered[k, individual_current] == 0:
                    #     print('crossover: ' + str(k) + ', ' + args_pop_selected[k, i].astype(str) + ' to ' + args_pop_crossovered[k, individual_current].astype(str))
                if args_pop_crossovered.shape[1] == 0:
                    args_pop_crossovered = args
                else:
                    args_pop_crossovered = np.insert(args_pop_crossovered, [1], args, axis=1)
                # print(args_pop_crossovered)
                # individual_current += 1
                # break

    #
    # IF THRE IS AN ERROR WITH 0 AS FIRST PARAMETER TRY TO PRINT args_pop_crossovered
    #

    # print(args_pop_crossovered)

    return args_pop_crossovered


def mutation(args_pop, mutation_probability):
    args_number = args_pop.shape[0]
    args_pop_mutated = np.zeros((args_number, args_pop.shape[1]), dtype=np.float16)

    for i in range(0, args_pop.shape[1]):
        for k in range(0, args_number):
            if k < args_number - 2:
                a_i_bits = np.binary_repr(int(args_pop[k, i] * 1000))
                a_i_bits_mutated = a_i_bits[0] + mutate_bits(a_i_bits[1:], mutation_probability)
                args_pop_mutated[k, i] = int(a_i_bits_mutated, 2) / 1000
            else:
                p_i_bits = np.binary_repr(int(args_pop[k, i]))
                p_i_bits_mutated = mutate_bits(p_i_bits, mutation_probability)
                args_pop_mutated[k, i] = int(p_i_bits_mutated, 2)
            # if args_pop_mutated[k, i] == 0:
            #     print('mutation: ' + str(k) + ', ' + args_pop[k, i].astype(str) + ' to ' + args_pop_mutated[k, i].astype(str))

    return args_pop_mutated