def train_model(layers,
inputs,
prices,
pop_size=150,
data_rotation=0,
w_mutation_rate = 0.05,
b_mutation_rate = 0.0,
mutation_scale = 0.3,
mutation_decay = 1.,
reporter=None):
# network parameters
network_params = {
'network': 'feedforward',
'input': inputs.shape[1],
'hidden': layers,
'output': 2
}
# build initial population
pop = Population(network_params,
pop_size,
mutation_scale,
w_mutation_rate,
b_mutation_rate,
mutation_decay,
socket_reporter=reporter)
g = 1 ###########
sample_size = 500 # request.json["sampleSize"]
while True:
if not g % data_rotation: ###########
ohlc, ta = data.get_rand_segment(sample_size) ##########
inputs, prices = data.get_training_segment() ##########
pop.evolve()
gen_best = pop.run((inputs, prices), fitness_callback=calc_overperformance)
g += 1
def train_model(layers,
inputs,
prices,
pop_size=150,
data_rotation=0,
w_mutation_rate=0.05,
b_mutation_rate=0.0,
mutation_scale=0.3,
mutation_decay=1.,
reporter=None):
# network parameters
network_params = {
'network': 'feedforward',
'input': inputs.shape[1],
'hidden': layers,
'output': 2
}
# build initial population
pop = Population(network_params,
pop_size,
mutation_scale,
w_mutation_rate,
b_mutation_rate,
mutation_decay,
socket_reporter=reporter)
g = 0
while True:
if g % data_rotation:
# need to rotate data here
pass
pop.evolve()
gen_best = pop.run((inputs, prices), fitness_callback=calculate_profit)
g += 1
# genetic parameters
pop_size = 4
w_mutation_rate = 0.05
b_mutation_rate = 0.0
mutation_scale = 0.3
mutation_decay = 0.995
generations = 500
# network parameters
network_params = {
'network': 'recurrent',
'timesteps': 4,
'input': inputs_train.shape[1],
'hidden': [16, 16, 16],
'output': 2
}
# build initial population
pop = Population(network_params, pop_size, mutation_scale, w_mutation_rate,
b_mutation_rate, mutation_decay)
# run for set number of generations
for g in range(generations):
pop.evolve(g)
gen_best = pop.run(inputs_train,
price_train,
fitness_callback=calculate_profit)
gen_best.save()
pop.test(inputs_test, price_test, fitness_callback=calculate_profit)
