/
train.py
196 lines (153 loc) · 6.14 KB
/
train.py
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
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
import _MultiNEAT as NEAT
from math import ceil, log2
from copy import deepcopy
import click
import logging
from tournament import Tournament, PlayOff
_action_vector_size = 3
_card_vector_size = 13 + 4
_red = "\x1b[31m"
_normal = "\x1b[0m"
_genome_store = "genome/{}-{}"
class Training:
LOGGER = logging.getLogger(name="Training")
def __init__(self, table_size, buy_in, min_denomination, tournament_rounds):
self.table_size = table_size
self.buy_in = buy_in
self.min_denomination = min_denomination
self.tournament_rounds = tournament_rounds
self.set_card_vector_size()
self.set_money_vector_size()
self.set_action_vector_size()
self.set_in_vector_size()
self.neat_parameters = NEAT.Parameters()
self.neat_parameters.DetectCompetetiveCoevolutionStagnation = True
self.neat_parameters.MutateAddNeuronProb = 0.1
self.neat_parameters.MutateAddLinkProb = 0.2
self.neat_parameters.CompatTreshold = 0.5
self.neat_parameters.DynamicCompatibility = True
self.neat_parameters.CompatTresholdModifier = 0.005
def set_card_vector_size(self):
self.card_vector_size = _card_vector_size
def set_money_vector_size(self):
normalized_max_chips = self.buy_in * self.table_size // self.min_denomination
self.money_vector_size = ceil(log2(normalized_max_chips))
def set_action_vector_size(self):
self.action_vector_size = _action_vector_size + self.money_vector_size
def set_in_vector_size(self):
size = 0
# hold cards
size += 2 * self.card_vector_size
# player bet
size += self.money_vector_size
# table cards
size += 5 * self.card_vector_size
# pot
size += self.money_vector_size
# current bet
size += self.money_vector_size
# player chips
size += self.table_size * self.card_vector_size
# turn indicator
size += self.table_size
# folded indicator
size += self.table_size
# last action
size += self.action_vector_size
self.in_vector_size = size
@staticmethod
def get_best_n_genomes(population, n):
num_genomes = population.NumGenomes()
genomes = [population.AccessGenomeByIndex(genome_index) for genome_index in range(num_genomes)]
genomes.sort(key=lambda genome: genome.GetFitness(), reverse=True)
return [deepcopy(genome) for genome in genomes[:n]]
def create_genome(self):
genome = NEAT.Genome(
0,
self.in_vector_size,
0,
self.action_vector_size,
True,
NEAT.ActivationFunction.UNSIGNED_SIGMOID,
NEAT.ActivationFunction.UNSIGNED_SIGMOID,
0,
self.neat_parameters,
0
)
return genome
def run(self):
logging.basicConfig(
format=_red + "[%(levelname)s][%(asctime)s][%(name)s] %(message)s" + _normal,
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.WARN
)
genome = self.create_genome()
population = NEAT.Population(genome, self.neat_parameters, True, 1.0, 0)
previous_best_genomes = None
current_best_genomes = None
difference = 0
self.LOGGER.warn("Population size: {}".format(population.NumGenomes()))
self.LOGGER.warn("{:>14s} {:>14s} {:>14s} {:>14s} {:>14s} {:>14s}".format(
"n_species",
"stagnation",
"mpc",
"search_mode",
"compat_thresh",
"improvement"
))
# run for 100 generations
for generation in range(200):
tournament = Tournament(
population,
self.table_size,
self.money_vector_size,
self.buy_in,
self.min_denomination,
self.tournament_rounds
)
tournament.play()
if current_best_genomes:
previous_best_genomes = current_best_genomes
current_best_genomes = self.get_best_n_genomes(population, self.table_size // 2)
if previous_best_genomes:
playoff = PlayOff(
current_best_genomes,
previous_best_genomes,
self.table_size,
self.money_vector_size,
self.buy_in,
self.min_denomination,
self.tournament_rounds
)
difference += playoff.play()
for index in range(population.NumGenomes()):
genome = population.AccessGenomeByIndex(index)
genome.SetFitness(genome.GetFitness() + difference)
self.LOGGER.warn("{:14d} {:14d} {:14.2f} {:>14s} {:14.3f} {:14.2f}".format(
len(population.Species),
population.GetStagnation(),
population.GetCurrentMPC(),
population.GetSearchMode().name,
population.Parameters.CompatTreshold,
difference
))
population.Epoch()
self.LOGGER.warn("Training complete")
return self.get_best_n_genomes(population, self.table_size)
def _save_genomes(genomes, genome_name):
genome_indices = range(len(genomes))
for index, genome in zip(genome_indices, genomes):
save_name = _genome_store.format(genome_name, index)
genome.Save(save_name)
@click.command()
@click.argument("genome_name", type=str)
@click.option("--table_size", default=8, help="Number of players round a table")
@click.option("--buy_in", default=8000, help="Number of chips each player starts with")
@click.option("--min_denomination", default=25, help="Minimum chip denomination")
@click.option("--tournament_rounds", default=10, help="Number of rounds per training tournament")
def main(genome_name, table_size=8, buy_in=8000, min_denomination=25, tournament_rounds=10):
training = Training(table_size, buy_in, min_denomination, tournament_rounds)
top_genomes = training.run()
_save_genomes(top_genomes, genome_name)
if __name__ == "__main__":
main()