def procedure(self):
best_ = np.Inf
best_sol = []
self.gd.build(self.greedyness, best_sol)
best_ = f(best_sol, self._adj_matrix)
for iter in range(self.Max):
new_sol = []
self.gd.build(self.greedyness, new_sol)
#Entrada que o shift 2 piora
#new_sol = [0, 2, 4, 1, 3, 0]
new_ = f(new_sol, self._adj_matrix)
aux_sol = [i for i in new_sol]
aux_ = new_
print('s=', aux_)
iterIls = 0
while iterIls < self.MaxIls:
#debug
#print((new_sol, new_))
new_sol = RVND(new_sol, self._adj_matrix, self.cache).execute()
new_ = f(new_sol, self._adj_matrix)
#debug
#print((new_sol, new_))
if aux_ > new_:
aux_ = new_
aux_sol = [i for i in new_sol]
iterIls = 0
#falta a pertubação
n = 5
x = (len(aux_sol)) // n
s = [aux_sol[1:x], aux_sol[(n - 1) * x:len(aux_sol) - 1]]
for i in range(1, n - 1):
a = choice([-1, 1])
s.append(aux_sol[i * x:(i + 1) * x][::a])
shuffle(new_sol)
a = []
for e in s:
a += e
new_sol = [0] + a + [0]
new_ = f(new_sol, self._adj_matrix)
print("p =", new_)
iterIls += 1
if best_ > aux_:
best_sol = aux_sol
best_ = aux_
Verbosity.show("{}/{} b={} a={}".format(iter + 1, self.Max, best_,
aux_))
return (best_sol, best_)
def one_epoch():
is_changed = 0
for x in dataset:
x1, x2 = x
correct = tools.f(x1, x2)
neuron_output = weights[0] + weights[1] * x1 + weights[2] * x2
result = tools.signum(neuron_output)
d = correct - result
if not d:
continue
weights[0] += tools.weight_delta(d, 1, weights[0])
weights[1] += tools.weight_delta(d, x1, weights[1])
weights[2] += tools.weight_delta(d, x2, weights[2])
tools.save_weights(weights)
is_changed = 1
return is_changed
def ml_features_normal(self, features):
from pyspark.ml.linalg import Vectors
from pyspark.ml.feature import Normalizer
from pyspark.ml.classification import NaiveBayes
from tools import f
features.foreach(print)
fea_df = features.map(lambda i: Row(**f(i))).toDF()
# fea_df.show()
normalizer = Normalizer().setInputCol('features').setOutputCol(
'norfeatures').setP(1.0)
norfea_df = normalizer.transform(fea_df)
# norfea_df.show()
train_dt, test_dt = norfea_df.randomSplit([0.8, 0.2])
nvby = NaiveBayes(modelType="multinomial", smoothing=0.1)
nvby_mod = nvby.fit(dataset=train_dt)
predictRDD = nvby_mod.transform(test_dt).rdd
count = predictRDD.count()
print(
predictRDD.map(lambda i: (i.label, i.prediction)).filter(
lambda i: i[0] == i[1]).count() / count)
os.makedirs(savepath)
except:
pass
if not Path.exists(
savepath / 'config.json'): # Only create json if it does not exist
with open(savepath / 'config.json', 'w') as fd:
json.dump(vars(args), fd)
# Generate data and create dataset
torch.manual_seed(args.dataset_seed)
np.random.seed(args.dataset_seed)
random.seed(args.dataset_seed)
X = (np.random.rand(10).reshape(-1, 1) - 1) / 2 # x between -0.5 and 0.
Y = f(X)
X = torch.from_numpy(X).type(torch.FloatTensor)
Y = torch.from_numpy(Y).type(torch.FloatTensor)
dataset = RegressionDataset(X, Y)
# Reproducibility
if args.seed is not None:
torch.manual_seed(args.seed)
np.random.seed(args.seed)
random.seed(args.seed)
net = MLP()
criterion = nn.MSELoss()
optimizer = optim.Adam(net.parameters(), lr=args.lr, weight_decay=args.wd)
import numpy
import sys
import tools
import TypeX
import TypeY
from matplotlib.pylab import *
I = numpy.arange(0.0,1.05,0.05)
f = []
string = "Plotting the firing rate of the neuron versus the amplitude"
string += "\r\n"
string += "of injecting step current. It may take several minutes ... "
string += "\r\n"
print string
for x in I:
# watch progress
print x
sys.stdout.flush()
f.append(tools.f(TypeX,x))
figure()
plot(I,f)
xlabel('Amplitude of Injecting step current (pA)')
ylabel('Firing rate (Hz)')
grid()
show()
###################################################
# try genetic algorithm ###########################
bests_percent, bests_min, bests_max = (10, 30, 500)
iter_count = 1000
resets_count = 100
mutations_count_percent = 60
generation = all_attributes.copy()
f_min = sys.float_info.max
opt_values = []
time_to_reset = int(iter_count / resets_count)
time_to_reset_initial = int(iter_count / resets_count)
for i in range(iter_count):
f_vals = [f(*attributes) for attributes in generation]
bests_count = int(np.ceil(len(f_vals) * bests_percent / 100))
if bests_count < bests_min:
bests_count = bests_min
elif bests_count > bests_max:
bests_count = bests_max
f_vals_bests_sorted = sorted(f_vals)[:bests_count]
f_min = f_vals_bests_sorted[0]
generation_bests = [
generation[f_vals.index(best)] for best in f_vals_bests_sorted
]
d = correct - result
if not d:
continue
weights[0] += tools.weight_delta(d, 1, weights[0])
weights[1] += tools.weight_delta(d, x1, weights[1])
weights[2] += tools.weight_delta(d, x2, weights[2])
tools.save_weights(weights)
is_changed = 1
return is_changed
i = 0
while one_epoch():
i += 1
print(i, 'epochs passed.')
# Тесты
for x in dataset:
x1, x2 = x
correct = tools.f(x1, x2)
neuron_output = weights[0] + weights[1] * x1 + weights[2] * x2
result = tools.signum(neuron_output)
if result != correct:
print('Error! Input values: ({}, {})'.format(*x))
break
os.makedirs(savepath)
except:
pass
if not Path.exists(
savepath / 'config.json'): # Only create json if it does not exist
with open(savepath / 'config.json', 'w') as fd:
json.dump(vars(args), fd)
# Generate data and create dataset
torch.manual_seed(args.dataset_seed)
np.random.seed(args.dataset_seed)
random.seed(args.dataset_seed)
X = (np.random.rand(10).reshape(-1, 1) - 1) / 2 # x between -0.5 and 0.
Y = f(X)
X = torch.from_numpy(X).type(torch.FloatTensor)
Y = torch.from_numpy(Y).type(torch.FloatTensor)
# Adding a single point at 0.35
nx = torch.tensor([[.25]]).float()
ny = torch.from_numpy(f(nx)).type(torch.FloatTensor)
X = torch.cat([X, nx])
Y = torch.cat([Y, ny])
dataset = RegressionDataset(X, Y)
# Reproducibility
if args.seed is not None:
torch.manual_seed(args.seed)
np.random.seed(args.seed)
