def mse_avg_table(train_x, train_y, test_x, test_y, file_name):
with open(file_name, 'w', newline='') as fileOut:
writer = csv.writer(fileOut,
delimiter=' ',
quotechar=' ',
quoting=csv.QUOTE_MINIMAL)
writer.writerow([
'hidden_neuron', 'avg_mse_train', 'st_d_train', 'avg_mse_test',
'st_d_test'
])
for i in range(
1, 42, 5
): # THIS RUNS FOR A LONG TIME!!! BECAUSE ITS 100 times training the neural network
errors_train = []
errors_test = []
for j in range(100):
print(str(j), '-ta iteracja')
network = rbf.RBF(i)
errors_train.append(
network.fit_two_stages(train_x, train_y, True))
errors_test.append(network.fit_two_stages(
test_x, test_y, True))
print(str(i) + ' neuron, errors train: ', errors_train)
print(str(i) + ' neuron, errors test: ', errors_test)
writer.writerow([
j,
round(np.average(errors_train), 4),
round(np.std(errors_train), 4),
round(np.average(errors_test), 4),
round(np.std(errors_test), 4)
])
def classification_test_table(data_test, data_train):
test_dataset = transform_data(data_test, len(data_test), 4, 3)
train_dataset = transform_data(data_train, len(data_train), 4, 3)
with open('results/classification/classification_effect_table.txt',
'w',
newline='') as fileOut:
writer = csv.writer(fileOut,
delimiter=' ',
quotechar=' ',
quoting=csv.QUOTE_MINIMAL)
writer.writerow([
'hidden_neurons', 'train_accuracy', 'st_d_train', 'test_accuracy',
'st_d_test'
])
for i in range(1, 42, 5):
acc_train = []
acc_test = []
for j in range(100):
network = rbf.RBF(i)
network.fit_two_stages(train_dataset[:, :-3],
train_dataset[:, -3:], False)
acc_train.append(accuracy_test(train_dataset, network))
acc_test.append(accuracy_test(test_dataset, network))
writer.writerow([
i + 1,
round(np.average(acc_train), 1),
round(np.std(acc_train), 2),
round(np.average(acc_test), 1),
round(np.std(acc_test), 2)
])
def task4(hidden_neurons, train_x, train_y, test_x):
time_jump = 5000
color = ['brown', 'darkslategray', 'b', 'r', 'c', 'm', 'lime', 'g']
plt.title("Approximation throughout epochs")
plt.xlabel('x')
plt.ylabel('f(x)')
plt.grid(alpha=.4, linestyle='--')
net = rbf.RBF(hidden_neurons)
net.fit_two_stages(train_x, train_y, True, 1, 1)
plt.plot(test_x, net.predict(test_x), c=color[0], label=str(1) + " epoch")
for j in range(1, 6):
net.fit_two_stages(train_x, train_y, True, 1, time_jump)
plt.plot(test_x,
net.predict(test_x),
c=color[j],
label=str(j * time_jump) + " epoch",
linewidth=1)
plt.scatter(train_data[:, 0],
train_data[:, 1],
c='k',
label="Train data",
marker='*',
alpha=0.2)
plt.legend(loc=2)
plt.savefig(
'results/approximation/approximation_throughout_epochs_data_1.png')
plt.clf()
def _fit(self): #train the network
try:
if self.radioButton_centrain.isChecked():
self.centskip = self.spinBox_centers.value()
if self.spinBox_centers.value() == 0: self.centskip = 1
self.centers = self.trainx[::self.centskip]
elif self.radioButton_rand.isChecked():
self.randnums = self.spinBox_rand.value()
self.centers = input.rancen(self.trainx, self.randnums)
elif self.radioButton.isChecked():
self.numlin = self.spinBox_equalcen.value()
if self.numlin == 0: self.numlin = 1
self.centers = input.lincen(self.trainx, self.numlin)
elif self.radioButton_2.isChecked():
if self.tempcents2 != []:
self.centers = self.tempcents2
#self.listWidget.addItem("Начало обучения")
if self.radioButton_3.isChecked():
self.NeuralNetwork = rbf.RBF(self.centers,
float(self.doubleSpinBox.value()),
self.func)
else:
try:
self.NeuralNetwork = rbf.TwoLayerRBF(
self.centers, float(self.doubleSpinBox.value()),
self.func)
except Exception as e:
self.warning.label.setText(str(e))
self.warning.show()
start = time.time()
self.NeuralNetwork.fit(self.trainx, self.trainy)
finish = time.time()
self.listWidget.addItem("Обучение завершено")
self.listWidget.addItem("Время обучения:")
tm = round(finish - start, 3)
if tm > 60:
tm /= 60
tm = str(tm) + "мин"
else:
tm = str(tm) + "c"
self.listWidget.addItem(tm)
self.listWidget.addItem("")
self.pushButton_saveweights.setEnabled(True)
self.pushButton.setEnabled(True)
if self.workx != []:
self.pushButton_predict.setEnabled(True)
except Exception as e:
self.warning.label.setText(str(e))
self.warning.show()
self.listWidget.addItem("Ошибка")
self.listWidget.addItem("")
def classification_accuracy(train_data, test_data, columns):
plt.title("Classification accuracy\n" + "(number of inputs " +
str(len(columns)) + ", columns " + str(columns) + ")")
plt.xlabel('number of radial neurons')
plt.ylabel('accuracy')
plt.grid(alpha=.4, linestyle='--')
x = [0, 1, 11, 21, 31, 41]
y = [0]
for n_neurons in range(1, 42, 10):
net = rbf.RBF(n_neurons)
net.fit_two_stages(train_data[:, :-3], train_data[:, -3:], False)
y.append(accuracy_test(test_data, net))
plt.plot(x, y, label='train data', alpha=0.8)
y = [0]
for n_neurons in range(1, 42, 10):
net = rbf.RBF(n_neurons)
net.fit_two_stages(test_data[:, :-3], test_data[:, -3:], False)
y.append(accuracy_test(test_data, net))
plt.plot(x, y, label='test data', alpha=0.8)
plt.legend()
plt.savefig('results/classification/classification_accuracy_' +
str(len(columns)) + '_inputs_c' + str(columns) + '.png')
plt.clf()
def select_ai(cl, data, targets, original_import):
""" Select an AI. Create in instance of the needed class and return it. """
ai = None
# MLP
if cl == "MLP":
ai = mlp.MLP(data,
targets,
hidden_nodes=MLP_HIDDEN_NODES,
beta=MLP_BETA,
momentum=MLP_MOMENTUM)
print "MLP created."
# RBF
elif cl == "RBF":
# Adjust target format for RBF: [win, draw, loss]
targets_rbf = ny.zeros((len(targets), 3))
for i in range(len(targets)):
win = 1 if targets[i] == WIN else 0
draw = 1 if targets[i] == DRAW else 0
loss = 1 if targets[i] == LOSS else 0
targets_rbf[i] = [win, draw, loss]
ai = rbf.RBF(data,
targets_rbf,
sigma=RBF_SIGMA,
rbfs_amount=RBF_NODES,
use_kmeans=RBF_KMEANS,
normalize=RBF_NORMALIZE)
print "RBF created."
# DTree
elif cl == "DTree":
data = []
for value in original_import:
zipped = zip(DATA_ATTRIBUTES, [datum for datum in value])
data.append(dict(zipped))
ai = dtree.DTree(data, DATA_ATTRIBUTES, DT_TARGET_ATTRIBUTE)
print "DTree created."
return ai
def _plot_radial_functions(hidden_neurons,
train_x,
train_y,
test_x,
backward=True,
coefficient=1.0,
l_rate=0.2,
b_rate=0.2,
epochs=1000):
network = rbf.RBF(hidden_neurons)
network.fit_two_stages(train_x, train_y, backward, coefficient, epochs,
l_rate, b_rate)
plt.plot(test_x, network.predict(test_x), label='network', linewidth=3)
G = network.calculate_matrix(test_x, coefficient)
for i in range(hidden_neurons):
for j in range(len(test_x)):
G[j][i] *= network.weights[i]
lab = str(i) + 'neuron'
plt.plot(test_x, G[:, i], label=lab, linewidth=2, alpha=0.7)
def decision_boundaries():
for i in range(3):
for j in range(i + 1, 4):
columns = [i, j]
plt.title("Distribution of points for inputs number " +
str(i + 1) + " and " + str(j + 1))
plt.xlabel('input ' + str(i + 1))
plt.ylabel('input ' + str(j + 1))
plt.grid(alpha=.4, linestyle='--')
iris = datasets.load_iris()
X = iris.data[:, columns]
y = iris.target
network = rbf.RBF(31)
network.fit_two_stages(X, y, False)
# Plotting decision regions
plot_decision_regions(X, y, clf=network, legend=2)
plt.savefig('results/classification/2222distribution_input_' +
str(i + 1) + '_and_' + str(j + 1))
plt.clf()
def _plot_5_predict(num,
step,
train_x,
train_y,
test_x,
backward=False,
coefficient=1.0,
l_rate=0.01,
b_rate=0.1,
epochs=10000):
for i in range(1, num, step):
network = rbf.RBF(i)
network.fit_two_stages(train_x, train_y, backward, coefficient, epochs,
l_rate, b_rate)
lab = str(i) + ' neuron(s)'
plt.plot(test_x,
network.predict(test_x),
label=lab,
linewidth=2,
alpha=0.7)
from sklearn.decomposition import PCA
print("PCA :", end=" ")
pca = PCA(0.8)
pca.fit(x_train)
x_train = pca.transform(x_train)
x_test = pca.transform(x_test)
print(pca.n_components_)
for i in range(len(y_train)):
y_train[i] = y_train[i] % 2
for i in range(len(y_test)):
y_test[i] = y_test[i] % 2
r = rbf.RBF(k=5, bias=True, kmeans=True)
start = time.process_time()
r.fit(x_train, y_train, showAcc=True, lr=0.01, epochs=13)
pos = -1
for i in range(0, 10000):
k = x_test[i]
pr = r.predictOne(k)
if pr == y_test[i]:
pos = i
break
if pos != -1:
k = x_test[pos]
pr = r.predictOne(k)
print(" Prediction : ", end=" ")
#from gym import wrappers
env = gym.make('MountainCar-v0')
#rewards = []
runs = 1
episodes = 300
numOfTilings = 8
alpha = 0.5
n = 1
nrbf_tiling = 8
# use optimistic initial value, so it's ok to set epsilon to 0
EPSILON = 0.0
centers = rbf.generate_grid_centers(nrbf_tiling)
print(centers)
widths = 0.1 * np.ones(len(centers))
rbfun = rbf.RBF(centers, widths, env.action_space.n)
fMap = rbf.Rbf_2D_Feature_Map(rbfun)
reward_fn = rbf.RbfReward(fMap,
np.array([-1. / len(centers) for _ in centers]),
env)
valueFunction = ValueFunction(alpha, numOfTilings)
for episode in range(0, episodes):
returns, states_visited, time = run_episode(env, valueFunction, n,
False, EPSILON, reward_fn)
#for s in states_visited:
# print(s, ",", reward_fn.get_reward(s))
print("-" * 10)
print(episode, returns, time)
#print('episode:', episode, "steps", step)
#pickle the controller (value function)
