def trainSingleNetwork(self, networkIndex, trainingInput, trainingOutput, testInput, testOutput):
trainingData = libfann.training_data()
trainingData.set_train_data(trainingInput, trainingOutput)
testData = libfann.training_data()
testData.set_train_data(testInput, testOutput)
mses = []
for i in range(0,300):
for j in range(0,10):
self.networks[networkIndex].train_epoch(trainingData)
mses.append(self.networks[networkIndex].test_data(testData))
if len(mses) > 10:
del mses[0]
neg = 0
for j in range(0,9):
if mses[j+1] - mses[j] < 0:
neg += 1
if neg < 2:
break
self.mses[networkIndex] = mses[9]
def test(file):
"""
excepts that <filename>.net and <filename>_test.data exists
Hardkoodattu kaksi tuloa ja yksi lähtö tulostuksiin
:param file: filename without extensions.
:return:
"""
net_file = file + '.net'
data_file = file + '_test.data'
testdata = libfann.training_data() # Luo olion
testdata.read_train_from_file(data_file) # Lukee testi materiaalin joka pitäisi olla eri kuin opetusmateriaali
inputs = testdata.get_input()
outputs = testdata.get_output()
ann = libfann.neural_net()
ann.create_from_file(net_file) # Lataa aikaisemmin luotu verkko
print("MSE ERROR : %.5f" %(ann.test_data(testdata))) # Ilmoittaa verkon virheen testidatalla
for i in range(len(inputs)): # Tulostaa testidatan läpi
result = ann.run(inputs[i])
print("Input: %.2f %.2f, Output %.4f, Excepted %.4f" %(inputs[i][0], inputs[i][1], result[0], outputs[i][0] ))
def train(self, train_data):
for sent in train_data.my_sents(self.name):
self.ids.add_sent(sent)
inputs = []
outputs = []
def add(vec, out):
inputs.append(self.vectorize(vec))
outputs.append([out])
def pollute(sent, p):
sent = sent[:]
for _ in range(int((len(sent) + 2) / 3)):
sent.insert(p, ':null:')
add(sent, self.LENIENCE)
def weight(sent):
def calc_weight(w): return pow(len(w), 3.0)
total_weight = 0.0
for word in sent:
total_weight += calc_weight(word)
for word in sent:
weight = 0 if word.startswith('{') else calc_weight(word)
add([word], weight / total_weight)
for sent in train_data.my_sents(self.name):
add(sent, 1.0)
weight(sent)
# Generate samples with extra unknown tokens unless
# the sentence is supposed to allow unknown tokens via the special :0
if not any(word[0] == ':' and word != ':' for word in sent):
pollute(sent, 0)
pollute(sent, len(sent))
for sent in train_data.other_sents(self.name):
add(sent, 0.0)
add([':null:'], 0.0)
add([], 0.0)
for sent in train_data.my_sents(self.name):
without_entities = sent[:]
for i, token in enumerate(without_entities):
if token.startswith('{'):
without_entities[i] = ':null:'
if without_entities != sent:
add(without_entities, 0.0)
inputs, outputs = resolve_conflicts(inputs, outputs)
train_data = fann.training_data()
train_data.set_train_data(inputs, outputs)
for _ in range(10):
self.configure_net()
self.net.train_on_data(train_data, 1000, 0, 0)
self.net.test_data(train_data)
if self.net.get_bit_fail() == 0:
break
def learn(self, episodes):
state_size = len(episodes[0].states[0])
# Create the model if needed
if self._model is None:
self._model = libfann.neural_net()
self._model.create_sparse_array(
1, (state_size, self.hidden_neurons, self.nb_actions))
self._model.randomize_weights(-0.1, 0.1)
self._model.set_activation_function_layer(libfann.GAUSSIAN, 1)
self._model.set_activation_function_layer(libfann.LINEAR, 2)
# Store the values of all the states encountered in all the episodes
states = []
values = []
for episode in episodes:
states.extend(episode.states)
values.extend(episode.values)
# Train for these values
data = libfann.training_data()
data.set_train_data(states, values)
self._model.train_on_data(data, 150, 50, 1e-5)
def train(self, train_data):
self.set_train_data(train_data)
hidden_layers = [self.hidden_neurons] * self.hidden_layers
layers = [self.train_data.num_input]
layers.extend(hidden_layers)
layers.append(self.train_data.num_output)
sys.stderr.write("Network layout:\n")
sys.stderr.write("* Neuron layers: %s\n" % layers)
sys.stderr.write("* Connection rate: %s\n" % self.connection_rate)
if self.training_algorithm not in ('TRAIN_RPROP',):
sys.stderr.write("* Learning rate: %s\n" % self.learning_rate)
sys.stderr.write("* Activation function for the hidden layers: %s\n" % self.activation_function_hidden)
sys.stderr.write("* Activation function for the output layer: %s\n" % self.activation_function_output)
sys.stderr.write("* Training algorithm: %s\n" % self.training_algorithm)
self.ann = libfann.neural_net()
self.ann.create_sparse_array(self.connection_rate, layers)
self.ann.set_learning_rate(self.learning_rate)
self.ann.set_activation_function_hidden(getattr(libfann, self.activation_function_hidden))
self.ann.set_activation_function_output(getattr(libfann, self.activation_function_output))
self.ann.set_training_algorithm(getattr(libfann, self.training_algorithm))
fann_train_data = libfann.training_data()
fann_train_data.set_train_data(self.train_data.get_input(), self.train_data.get_output())
self.ann.train_on_data(fann_train_data, self.epochs, self.iterations_between_reports, self.desired_error)
return self.ann
def fit(self, X, y):
self._create_network()
x_train = libfann.training_data()
if len(y.shape) == 1 and y.shape[0] > 0:
""" fann requires a row vector"""
y = y[:, np.newaxis]
x_train.set_train_data(X, y)
self.ann.train_on_data(x_train, self.epoch, self.show, self.desired_error)
def fit(self, X, y):
self._create_network()
x_train = libfann.training_data()
if len(y.shape) == 1 and y.shape[0] > 0:
""" fann requires a row vector"""
y = y[:, np.newaxis]
x_train.set_train_data(X, y)
self.ann.train_on_data(x_train, self.epoch, self.show,
self.desired_error)
def test(self, test_data):
self.set_test_data(test_data)
fann_test_data = libfann.training_data()
fann_test_data.set_train_data(self.test_data.get_input(), self.test_data.get_output())
self.ann.reset_MSE()
self.ann.test_data(fann_test_data)
return self.ann.get_MSE()
def fann_train(self, train_data, train_tar, net): data = libfann.training_data() data.set_train_data(train_data, train_tar) input_layer_num = data.num_input_train_data() output_layer_num = data.num_output_train_data() hidden_layer_num = int(float(input_layer_num + output_layer_num) / 2.0) print "Training Network With Params:\nInput: %s, Hidden: %s, Output: %s" %(input_layer_num, hidden_layer_num, output_layer_num) net.create_sparse_array(1, (input_layer_num, hidden_layer_num, output_layer_num)) net.train_on_data(data, self.epocs, self.disp_round, self.error_rate) return net
def fann_train(self, train_data, train_tar, net):
data = libfann.training_data()
data.set_train_data(train_data, train_tar)
input_layer_num = data.num_input_train_data()
output_layer_num = data.num_output_train_data()
hidden_layer_num = int(float(input_layer_num + output_layer_num) / 2.0)
print "Training Network With Params:\nInput: %s, Hidden: %s, Output: %s" % (
input_layer_num, hidden_layer_num, output_layer_num)
net.create_sparse_array(
1, (input_layer_num, hidden_layer_num, output_layer_num))
net.train_on_data(data, self.epocs, self.disp_round, self.error_rate)
return net
def train(self, train_data):
for sent in train_data.my_sents(self.name):
self.ids.add_sent(sent)
inputs = []
outputs = []
def add(vec, out):
inputs.append(self.vectorize(vec))
outputs.append([out])
def pollute(sent, p):
sent = sent[:]
for _ in range(int((len(sent) + 2) / 3)):
sent.insert(p, ':null:')
add(sent, self.LENIENCE)
def weight(sent):
def calc_weight(w):
return pow(len(w), 3.0)
total_weight = 0.0
for word in sent:
total_weight += calc_weight(word)
for word in sent:
weight = 0 if word.startswith('{') else calc_weight(word)
add([word], weight / total_weight)
for sent in train_data.my_sents(self.name):
add(sent, 1.0)
weight(sent)
if not any(word[0] == ':' for word in sent):
pollute(sent, 0)
pollute(sent, len(sent))
for sent in train_data.other_sents(self.name):
add(sent, 0.0)
add([], 0.0)
inputs, outputs = resolve_conflicts(inputs, outputs)
train_data = fann.training_data()
train_data.set_train_data(inputs, outputs)
for _ in range(10):
self.configure_net()
self.net.train_on_data(train_data, 1000, 0, 0)
self.net.test_data(train_data)
if self.net.get_bit_fail() == 0:
break
def train(self, train_data):
for sent in train_data.my_sents(self.intent_name):
if self.token in sent:
for i in range(
sent.index(self.token) + self.dir, self.get_end(sent),
self.dir):
if sent[i][0] != '{':
self.ids.add_token(sent[i])
inputs, outputs = [], []
def pollute(sent, i, out_val):
"""Simulates multiple token words in adjacent entities"""
for j, check_token in enumerate(sent):
d = j - i
if int(d > 0) - int(
d < 0) == self.dir and check_token.startswith('{'):
for pol_len in range(1, 4):
s = sent[:j] + [':0'] * pol_len + sent[j + 1:]
p = i + (pol_len - 1) * int(self.dir < 0)
inputs.append(self.vectorize(s, p))
outputs.append([out_val])
def add_sents(sents, out_fn):
for sent in sents:
for i, token in enumerate(sent):
out_val = out_fn(token)
inputs.append(self.vectorize(sent, i))
outputs.append([out_val])
if out_val == 1.0:
pollute(sent, i, 1.0)
add_sents(train_data.my_sents(self.intent_name),
lambda x: float(x == self.token))
add_sents(train_data.other_sents(self.intent_name), lambda x: 0.0)
inputs, outputs = resolve_conflicts(inputs, outputs)
data = fann.training_data()
data.set_train_data(inputs, outputs)
for _ in range(10):
self.configure_net()
self.net.train_on_data(data, 1000, 0, 0)
self.net.test_data(data)
if self.net.get_bit_fail() == 0:
break
def train_on_minibatch(self):
n = min(300, len(self.learnbuffer))
if n < 300: return
minibatch = random.sample(self.learnbuffer, n)
inputs = []
outputs = []
for oldstate, action, newstate, reward in minibatch:
diff, val = self.value_update(oldstate, action, newstate, reward)
val[action] += diff
inputs += [[oldstate[0] / 7., oldstate[1] / 5.]]
outputs += [val]
#print("training minibatch of size %i:\n%s\n%s\n\n"%(n, str(inputs), str(outputs)))
training_data = libfann.training_data()
training_data.set_train_data(inputs, outputs)
self.NN.train_epoch(training_data)
def train_on_minibatch(self):
n = min(300, len(self.learnbuffer))
if n < 300: return
minibatch = random.sample(self.learnbuffer, n)
inputs = []
outputs = []
for oldstate, action, newstate, reward in minibatch:
diff, val = self.value_update(oldstate, action, newstate, reward)
val[action] += diff
inputs += [ [oldstate[0]/7., oldstate[1]/5.] ]
outputs += [ val ]
#print("training minibatch of size %i:\n%s\n%s\n\n"%(n, str(inputs), str(outputs)))
training_data = libfann.training_data()
training_data.set_train_data(inputs, outputs)
self.NN.train_epoch(training_data)
def XY_to_fann_train_data(X, Y):
if len(X) != len(Y):
raise ValueError("X and Y must have the same number of lines.")
train_data = libfann.training_data()
if len(X):
dim_X, dim_Y = len(X[0]), len(Y[0])
tmp = tempfile.NamedTemporaryFile(delete=False)
with tmp:
tmp.write("%d %d %d\n"%(len(X), dim_X, dim_Y))
for i in xrange(len(X)):
for line in [ X[i], Y[i] ]:
tmp.write("%s\n"% ' '.join( str(float(val)) for val in line ))
train_data.read_train_from_file(tmp.name)
tmp.unlink(tmp.name)
return train_data
def learn(train_file):
"""
:param train_file: filename without extensions.
:return:
method will create nn called <filename>.net
and except that <filename>.data exists
Kaikki arvot on kovakoodattu tällä hetkellä. Voisi olla vapaaehtoisia parametrejä.
Hidden layerien määrän hallintaan pitää keksiä jokin juttu.
"""
net_file = train_file + '.net'
data_file = train_file + '_train.data'
connection_rate = 1
learning_rate = 0.5 # learning rate ei saa olla liian suuri, toisaalta liian pienellä oppiminen kestää kauan
num_input = 2
num_hidden = 4
num_output = 1
desired_error = 0.00005
max_iterations = 100000
iterations_between_reports = 10
ann = libfann.neural_net()
# Lue tiedosto
trainindata = libfann.training_data()
trainindata.read_train_from_file(data_file)
# Luo verkon
ann.create_sparse_array(connection_rate, (num_input, num_hidden, num_hidden, num_output))
ann.set_learning_rate(learning_rate)
ann.set_activation_function_output(libfann.SIGMOID_SYMMETRIC_STEPWISE) # Aktivointi funktio
ann.set_training_algorithm(libfann.TRAIN_INCREMENTAL) # Oppimis algoritmi
ann.train_on_data(trainindata, max_iterations, iterations_between_reports, desired_error)
ann.save(net_file)
def createDistanceTestFile(filename, length, max=0.707, trainTest=False):
"""
Esimerkki funktio datan luonnista. Tein ihan itse
:param filename: desired filename without exteensions
:param length: How many input, output pairs
:return:
"""
inputs = []
outputs = []
for i in range(length):
if trainTest:
p = random.randrange(1)
if p == 0:
a = random.uniform(0, 0.2)
else:
a = random.uniform(0.5, max)
p = random.randrange(1)
if p == 0:
b = random.uniform(0, 0.2)
else:
b = random.uniform(0.5, max)
else:
a = random.uniform(0, max)
b = random.uniform(0, max)
c = math.sqrt(math.pow(a, 2) + math.pow(b, 2))
inputs.append([a, b])
outputs.append([c])
data = libfann.training_data()
data.set_train_data(inputs, outputs)
#data.scale_input_train_data(-1.0, 1.0)
#data.scale_output_train_data(-1.0, 1.0)
data.save_train(filename)
def trainANN(features, labels, connRate, hidNodes, error, binary):
"""
Train the neural network using the given training data and
parameters. Returns a fully trained ANN.
"""
# Organize ANN parameters
connection_rate = connRate
num_input = 72
num_hidden = hidNodes
num_output = 3
desired_error = error
max_iterations = 100000
# Print out two reports for every ANN
iterations_between_reports = 50000
# Binarize labels as it is necessary for ANN
labels = binary.fit_transform(labels)
# Cast numpy to python list
annFeatures = features.tolist()
annLabels = labels.tolist()
# Create an ANN training data instance and set data
training = libfann.training_data()
training.set_train_data(annFeatures, annLabels)
ann = libfann.neural_net()
ann.create_sparse_array(connection_rate,
(num_input, num_hidden, num_output))
# Train the ANN
ann.train_on_data(training, max_iterations, iterations_between_reports,
desired_error)
return ann
def trainANN( features, labels, connRate, hidNodes, error, binary ): """ Train the neural network using the given training data and parameters. Returns a fully trained ANN. """ # Organize ANN parameters connection_rate = connRate num_input = 72 num_hidden = hidNodes num_output = 3 desired_error = error max_iterations = 50000 # Print out two reports for every ANN iterations_between_reports = 50000 # Binarize labels as it is necessary for ANN labels = binary.fit_transform( labels ) # Cast numpy to python list annFeatures = features.tolist() annLabels = labels.tolist() # Create an ANN training data instance and set data training = libfann.training_data() training.set_train_data( annFeatures, annLabels ) ann = libfann.neural_net() ann.create_sparse_array( connection_rate, (num_input, num_hidden, num_output) ) # Train the ANN ann.train_on_data( training, max_iterations, iterations_between_reports, desired_error ) return ann
def learn(self, episodes):
state_size = len(episodes[0].states[0])
# Create the model if needed
if self._model is None:
self._model = libfann.neural_net()
self._model.create_sparse_array(1, (state_size, self.hidden_neurons, self.nb_actions))
self._model.randomize_weights(-0.1, 0.1)
self._model.set_activation_function_layer(libfann.GAUSSIAN, 1)
self._model.set_activation_function_layer(libfann.LINEAR, 2)
# Store the values of all the states encountered in all the episodes
states = []
values = []
for episode in episodes:
states.extend(episode.states)
values.extend(episode.values)
# Train for these values
data = libfann.training_data()
data.set_train_data(states, values)
self._model.train_on_data(data, 150, 50, 1e-5)
import os
from fann2 import libfann as fann
import cv2 as cv
ENG_ALPHABET = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
outputs = []
inputs = []
for letter in os.listdir('./letters/'):
print(letter)
for letter_img in os.listdir('./letters/' + letter):
img = cv.imread(os.path.join('./letters', letter, letter_img))
img = cv.resize(img, (35, 35))
img = cv.threshold(img, 0, 255, cv.THRESH_BINARY)[1]
img_bin = []
for x in img:
for y in x:
if sum(y) == 0:
img_bin.append(1)
else:
img_bin.append(0)
result_array = ([0] * 26)
result_array[ENG_ALPHABET.index(letter)] = 1
outputs.append(result_array)
inputs.append(img_bin)
td = fann.training_data()
td.set_train_data(inputs, outputs)
td.save_train('./train.dat')
def train(self):
print("Retraining the ANN using specified predictors.\n")
# Collect the ratings by the predictors
R_T = []
for p in self.predictors:
g = pandas.read_csv(
ROOT + '/results/blend/' + p + '_probe',
header=None,
names=['rating']
)['rating'].tolist()
# Convert the list to a numpy array
g = np.array(g, dtype=float)
# Subtract out the mean rating of p
# g -= np.sum(g) / len(g)
g /= 5
R_T.append(g)
# Convert the predictor ratings into a matrix to transpose it
R_T = np.matrix(R_T)
R = R_T.T
# Get it back in list of list form, each inner list is the set of
# predictions for a single user/movie combination
self.inputs = R.tolist()
# Get a list of the actual probe ratings
probe = pandas.read_csv(
ROOT + '/data/parsed/probe_ratings.dta',
header=None,
names=['rating']
)['rating'].tolist()
# Convert it into a numpy array
probe = np.array(probe, dtype=float)
# Subtract out the mean rating of probe
# probe -= np.sum(probe) / len(probe)
probe /= 5
# Create a list for each item of the list.
outputs = [[x] for x in probe.tolist()]
# Parameters for neural network
# Originally, 5e-4 and 3e-7
# 5e-5 and 1.5e-8 worked well and was still decreasing in RMSE
# until 3000 epochs
ANN_LEARNRATE = 0.0005
ANN_LEARNRATE_STEP = 0.00000015
NUM_INPUT = len(self.inputs[0])
NUM_OUTPUT = len(outputs[0]) # Should be 1
NUM_HIDDEN = 19
# Create an artificial neural network.
self.ann = libfann.neural_net()
self.ann.set_learning_rate(ANN_LEARNRATE)
self.ann.set_activation_function_output(libfann.SIGMOID_SYMMETRIC)
# ann.set_activation_function_output(libfann.LINEAR)
# ann.set_activation_function_output(libfann.SIGMOID)
self.ann.set_training_algorithm(libfann.TRAIN_INCREMENTAL)
self.ann.create_standard_array((NUM_INPUT, NUM_HIDDEN, NUM_OUTPUT))
# Set up the training data for the neural network.
train_data = libfann.training_data()
train_data.set_train_data(self.inputs, outputs)
# Train the network
MAX_ITER = 3000
REPORT_SPACING = 25
# DESIRED_E = 0.0001
# Sanity check
assert ANN_LEARNRATE - (MAX_ITER * ANN_LEARNRATE_STEP) > 0
# ann.train_on_data(train_data, MAX_ITER, REPORT_SPACING, DESIRED_E)
for i in range(MAX_ITER):
learn_rate = ANN_LEARNRATE - (i * ANN_LEARNRATE_STEP)
self.ann.set_learning_rate(learn_rate)
self.ann.train_epoch(train_data)
epoch = i+1
if epoch == 1:
rmse = np.sqrt(25 * self.ann.get_MSE())
best_rmse = rmse
best_epoch = 1
if epoch % REPORT_SPACING == 0 or epoch == 1:
rmse = np.sqrt(25 * self.ann.get_MSE())
# Keep track of the best output so far
if rmse < best_rmse:
best_rmse = rmse
best_epoch = epoch
print("Finished epoch {0} with RMSE {1}.".format(epoch, rmse))
# TODO: Also need to predict probe ratings if we want to use
# these blends in a further linear blend.
print("Predicting qual ratings for epoch {0}.".format(epoch))
self.predict(suffix=str(epoch))
print(" - Done predicting qual ratings.")
print("Predicting probe ratings for epoch {0}.".format(epoch))
self.probe_predict(suffix=str(epoch))
print(" - Done predicting probe ratings.\n")
# Save the network
netfile = ROOT + '/results/ann.net'
print("Neural network saved to {0}\n".format(netfile))
self.ann.save(netfile)
# Print best RMSE & epoch
print("Best epoch RMSE was {0}.".format(best_rmse))
print(" - Corresponds to epoch {0}".format(best_epoch))
def train_and_test(results, iterations, output_dir, name):
data = read_metrics(results)
split = int(len(data) * 0.8)
num_input = len(data[0]) - 1
random.shuffle(data)
# # Plot train data output distribution
# outputs_train = sorted([row[num_input] for row in data[:split]])
# outputs_test = sorted([row[num_input] for row in data[split:]])
# fig = plt.figure(figsize=(6, 4))
# plt.plot(outputs_train)
# plt.ylim([0, 2])
# plt.show()
# fig = plt.figure(figsize=(6, 4))
# plt.plot(outputs_test)
# plt.ylim([0, 2])
# plt.show()
# return None
train_data_file = _get_ann_data_file(data[:split])
test_data_file = _get_ann_data_file(data[split:])
ann = libfann.neural_net()
ann.create_standard_array([num_input, 20, 5, 1])
# ann.set_activation_function_layer(libfann.SIGMOID, 1)
# ann.set_activation_function_layer(libfann.SIGMOID, 2)
ann.set_activation_steepness_hidden(1.0)
# Train
# TEST TEST TEST
# train_data = libfann.training_data()
# train_data.read_train_from_file(train_data_file.name)
# ann.init_weights(train_data)
# ann.set_learning_rate(1.0)
# ann.set_training_algorithm(libfann.TRAIN_INCREMENTAL)
ann.train_on_file(train_data_file.name, iterations, 10, 0.0)
# Save results
ann.save(os.path.join(output_dir, name + '.net'))
shutil.copy(train_data_file.name, os.path.join(output_dir, name + '_train.data'))
shutil.copy(test_data_file.name, os.path.join(output_dir, name + '_test.data'))
# Test
print "train MSE: " + str(ann.get_MSE())
test_data = libfann.training_data()
test_data.read_train_from_file(test_data_file.name)
ann.reset_MSE()
ann.test_data(test_data)
print "test MSE: " + str(ann.get_MSE())
test_data.destroy_train()
outputs = sorted([
(ann.run(row[:num_input]), row[num_input])
for row in data[split:]
], key=lambda p: p[1])
fig = plt.figure(figsize=(6, 4))
plt.plot([p[0] for p in outputs])
plt.plot([p[1] for p in outputs])
plt.ylim([0, 1.2])
plt.show()
train_data_file.close()
test_data_file.close()
return None
# res.append([data[i+N+1]])
indvec = range(len(data))
random.shuffle(indvec)
#print(len(indvec))
#print("indvec",indvec)
vec_train = [data[i] for i in indvec[:int(len(data)*0.8)]]
res_train = [ext[i] for i in indvec[:int(len(data)*0.8)]]
vec_test = [data[i] for i in indvec[int(len(data)*0.8):]]
res_test = [ext[i] for i in indvec[int(len(data)*0.8):]]
#print(len(vec_train),len(res_train))"
#print vec_train
#print len(data)
#print len(ext)
tr_data = libfann.training_data()
tr_data.set_train_data(vec_train,res_train)
#print(vec_train)
ann.train_on_data(tr_data,max_iter,iter_betw_repor,error)
thr = 0.0
auc = 0.0
last_fpr = 0
last_tpr = 0
max_acc = 0.0
for p in range(0,1001):
#0.1....0.9 postrROP
(fp,tp,fn,tn) = (0,0,0,0) # hypothises is this vector is from malicious process
#run on test sample
for i in range(len(vec_test)):
bbb = ann.run(vec_test[i])
if res_test[i][0] == 1.0:
from fann2 import libfann as pyfann
connection_rate = .5
learning_rate = 0.5
desired_error = 0.00000001
max_iterations = 3500
iterations_between_reports = 50
training = pyfann.training_data()
training.read_train_from_file("/home/ubuntu/svidur/data_train.data")
num_input = len(training.get_input()[0])
num_neurons_hidden_1 = 24
num_neurons_hidden_2 = 7
num_output = len(training.get_output()[0])
ann = pyfann.neural_net()
#ann.create_sparse_array(connection_rate, (num_input,num_neurons_hidden_1,num_neurons_hidden_2, num_output))
ann.create_standard_array((num_input,num_neurons_hidden_1,num_neurons_hidden_2,num_output))
ann.set_activation_function_hidden(pyfann.ELLIOT_SYMMETRIC)
ann.set_activation_function_output(pyfann.ELLIOT_SYMMETRIC)
ann.set_training_algorithm(pyfann.TRAIN_BATCH)
ann.set_train_error_function(pyfann.ERRORFUNC_LINEAR)
ann.train_on_data(training, max_iterations, iterations_between_reports, desired_error)
print "Testing network"
test_data = pyfann.training_data()
# res.append([data[i+N+1]])
indvec = range(len(data))
random.shuffle(indvec)
#print(len(indvec))
#print("indvec",indvec)
vec_train = [data[i] for i in indvec[:int(len(data) * 0.8)]]
res_train = [ext[i] for i in indvec[:int(len(data) * 0.8)]]
vec_test = [data[i] for i in indvec[int(len(data) * 0.8):]]
res_test = [ext[i] for i in indvec[int(len(data) * 0.8):]]
#print(len(vec_train),len(res_train))"
#print vec_train
#print len(data)
#print len(ext)
tr_data = libfann.training_data()
tr_data.set_train_data(vec_train, res_train)
#print(vec_train)
ann.train_on_data(tr_data, max_iter, iter_betw_repor, error)
thr = 0.0
auc = 0.0
last_fpr = 0
last_tpr = 0
max_acc = 0.0
for p in range(0, 1001):
#0.1....0.9 postrROP
(fp, tp, fn,
tn) = (0, 0, 0, 0) # hypothises is this vector is from malicious process
#run on test sample
for i in range(len(vec_test)):
bbb = ann.run(vec_test[i])
def get_train_data(self, trainfile):
traindata = libfann.training_data()
traindata.read_train_from_file(trainfile)
return traindata
connection_rate = 1
learning_rate = 0.7
num_neurons_hidden = 100
desired_error = 0.0001
max_iterations = 100
iterations_between_reports = 1
early_stopping_threshold = 3
break_on_early_stopping = False
ann = libfann.neural_net()
train_data = libfann.training_data()
train_data.read_train_from_file("data/dev-kmeans-10-pca.data")
# train_data.scale_input_train_data(0, 1)
# ann.create_from_file("minimal.net")
ann.create_sparse_array(connection_rate, (train_data.num_input_train_data(), 40, 20, 10, train_data.num_output_train_data()))
ann.set_learning_rate(learning_rate)
ann.set_activation_function_output(libfann.SIGMOID)
test_data = libfann.training_data()
test_data.read_train_from_file("data/test-kmeans-10-pca.data")
# test_data.scale_input_train_data(0, 1)
count = 0