def nextGeneration(self, tetronimoes):
scores = [(strategy, strategy.score(tetronimoes))
for strategy in self.strategies]
scores.sort(reverse=True, key=lambda x: x[1])
# keep best 25%
self.strategies = []
for i in range(int(self.size / 4)):
self.strategies.append(scores[i][0])
# generate 10% new data for the gene pool
for _ in range(int(self.size / 10)):
self.strategies.append(
Strategy.generateRandomStrategy(self.functions))
# for 10% of the data, randomly select 20% of the data, swap a weight in best 2
for _ in range(int(self.size / 10)):
subset = random.sample(scores, int(self.size / 5))
subset.sort(reverse=True, key=lambda x: x[1])
index = random.sample(
[i for i in range(len(subset[0][0].weights))], 2)
subset[0][0].weights[index[0]], subset[0][0].weights[
index[1]] = subset[0][0].weights[
index[1]], subset[0][0].weights[index[0]]
self.strategies.append(
Strategy(self.functions, subset[0][0].weights))
# fill the rest of the strategies with evolved data
# randomly select 20% of the data, average the best 2
while len(self.strategies) < self.size:
subset = []
if self.size >= 10:
subset = random.sample(scores, int(self.size / 5))
else:
subset = random.sample(scores, 2)
subset.sort(reverse=True, key=lambda x: x[1])
weights = [
subset[0][0].weights[f] + subset[1][0].weights[f]
for f in range(len(self.functions))
]
weightnorm = pow(sum(w**2 for w in weights), 0.5)
weights = [w / weightnorm for w in weights]
self.strategies.append(Strategy(self.functions, weights))
#replace repeating stratagies with random weights
self.strategies.sort(reverse=True, key=lambda x: x.weights[0])
for i in range(self.size):
if i != 0 and self.strategies[i - 1].weights == self.strategies[i]:
self.strategies[i - 1] = generateRandomStrategy(self.functions)
return scores, sum(score for (_, score) in scores) / len(scores)
def __init__(self, functions, input, size=100):
assert (size >= 2)
self.functions = functions
self.size = size
self.strategies = []
if input:
with open(input, 'r') as inputWeights:
reader = csv.reader(inputWeights)
for row in reader:
weights = list(map(float, row))
if len(functions) != len(weights):
sys.exit(
"input file weight list is not the same length as the functions list used"
)
self.strategies.append(Strategy(functions, weights))
self.size = len(self.strategies)
else:
for _ in range(size):
self.strategies.append(
Strategy.generateRandomStrategy(functions))
