def prevertemperatura(params):
inputs = []
inputs.append(params['chuva'])
inputs.append(params['hora'])
inputs.append(params['tanterior'])
inputs.append(params['tatual'])
inputs.append(params['ranterior'])
inputs.append(params['ratual'])
inputs.append(params['poanterior'])
inputs.append(params['poatual'])
temperaturas = []
temperaturas.append(float(params['tatual']))
with open('A702_normalizado.pickle', 'r') as arquivo:
arquivoinput = pickle.load(arquivo)
inputNorm = nl.tool.Norm(arquivoinput)
inputNormTarget = inputNorm(inputs)
for i in range(1, 7):
with open('A702_t+' + str(i) + '.pickle', 'r') as arquivo:
output = pickle.load(arquivo)
net = nl.load('t+' + str(i) + '.net')
out = net.sim(inputNormTarget)
outputNorm = nl.tool.Norm(output)
temperaturas.append(outputNorm.renorm(out)[0][0])
return temperaturas
def neuro_predict(x0, y0, y, name):
import neurolab as nl
import numpy as np
size = len(y)
inp = np.concatenate((x0, y0), axis=1)
tar = y.reshape(size,1)
# data preprocessing
mean_tar = tar.mean(axis=0)[0]
std_tar = tar.std(axis=0)[0]
import os
cwd = os.getcwd()
sub_dir = cwd + '/training_models/neuro_network'
filename = 'model_' + name
#net = nl.load( os.path.join(sub_dir, filename) )
net = nl.load(filename)
out = net.sim(inp).reshape(size)
out = out * 2 * std_tar + mean_tar
return out
def preverradiacao(params):
inputs = []
inputs.append(params['chuva'])
inputs.append(params['hora'])
inputs.append(params['tanterior'])
inputs.append(params['tatual'])
inputs.append(params['ranterior'])
inputs.append(params['ratual'])
inputs.append(params['poanterior'])
inputs.append(params['poatual'])
radiacaos = []
radiacaos.append(float(params['ratual']))
with open('A702_normalizado.pickle', 'r') as arquivo:
arquivoinput = pickle.load(arquivo)
inputNorm = nl.tool.Norm(arquivoinput)
inputNormTarget = inputNorm(inputs)
for i in range(1, 7):
with open('A702_radiacao+' + str(i) + '.pickle', 'r') as arquivo:
output = pickle.load(arquivo)
net = nl.load('radiacao+' + str(i) + '.net')
out = net.sim(inputNormTarget)
outputNorm = nl.tool.Norm(output)
saida = outputNorm.renorm(out)[0][0]
if saida < 1:
saida = 0
radiacaos.append(saida)
return radiacaos
def setNN(input, output, path, neurons, epochs):
print("\tResilient backpropagation network with {0} neurons in hidden layer and {1} epochs"\
.format(neurons, epochs))
XInput, YOutput, XTest, YTest = splitData(input, output)
if os.path.isfile(path):
print("\n\tLoading network from " + path)
return nl.load(path)
net = nl.net.newff([[-1, 1]] * len(XInput[0]), [neurons, 16])
net.trainf = nl.net.train.train_rprop
net.init()
result = net.train(XInput, YOutput, epochs=epochs, show=10, goal=0.0001)
print("\tSaving network in " + path)
net.save(path)
res = net.sim(XInput)
plot.plot(XInput, YOutput)
#how many good classifications
true = 0
for i in range(0, len(res)):
for j in range(0, 16):
if (res[i][j] == max(res[i])):
if (YOutput[i][j] == 1.0):
true += 1
return net
def neuro_predict(x0, y0, y, name):
import neurolab as nl
import numpy as np
size = len(y)
inp = np.concatenate((x0, y0), axis=1)
tar = y.reshape(size, 1)
# data preprocessing
mean_tar = tar.mean(axis=0)[0]
std_tar = tar.std(axis=0)[0]
import os
cwd = os.getcwd()
sub_dir = cwd + '/training_models/neuro_network'
filename = 'model_' + name
#net = nl.load( os.path.join(sub_dir, filename) )
net = nl.load(filename)
out = net.sim(inp).reshape(size)
out = out * 2 * std_tar + mean_tar
return out
def red_neuronal(): #Parametros de entrada para el tamaño del tablero
if os.path.isfile('../Pentominos/redes/red-50k-onlymax.net') == False:
funcion_activacion = trans.LogSig()
red = net.newff(minmax=[[0, 1]] * 63,
size=[7, 7, 63],
transf=[funcion_activacion] * 3)
red.reset()
entrada, objetivo, prueba = get_entrada_objetivo()
#Sesgos y pesos iniciales
np.random.seed(3287426346)
# red.reset()
for capa in red.layers:
capa.initf = init.init_zeros
red.init()
red.trainf = train.train_gd
red.errorf = error.MAE()
print("Comienza el entrenamiento")
print("Net.ci: " + str(red.ci))
red.train(entrada, objetivo, lr=0.1, epochs=50000, show=500, goal=0.01)
red.sim(prueba)
red.save('../Pentominos/redes/red-50k-onlymax.net')
# return red.sim(prueba), prueba
else:
red = nl.load('../Pentominos/redes/red-50k-onlymax.net')
return red
def create_nn(input_data,
output_data,
path,
hidden_layers=8,
epochs=500,
goal=0.5):
if os.path.isfile(path):
print("\n\tLoading network from " + path)
net = nl.load(path)
res = net.sim(input_data)
return net, res
# init the neural network
net = nl.net.newff([[-100, 100]] * len(input_data[0]),
[hidden_layers, len(output_data[0])])
net.trainf = nl.net.train.train_rprop
net.init()
# train the neural network
net.train(input_data, output_data, epochs=epochs, show=10, goal=goal)
# save network to a file
print("\tSaving network in " + path)
net.save(path)
# get the results
res = net.sim(input_data)
return net, res
def pegarRedeSalva(self, widget):
try:
dialog = Gtk.FileChooserDialog("Por favor, escolha um arquivo .net",
None,
Gtk.FileChooserAction.OPEN,
(Gtk.STOCK_CANCEL, Gtk.ResponseType.CANCEL,
Gtk.STOCK_OPEN, Gtk.ResponseType.OK))
response = dialog.run()
if response == Gtk.ResponseType.OK:
#carrega o arquivo
self.net = neurolab.load(dialog.get_filename())
#extrai e coloca nos liststores
numTargets = [float(self.targets[i][j])
for i in xrange(len(self.targets)) for j in xrange(1)]
self._setListStore(self.inputs, numTargets, True)
self.storeRA.append([str(self.net.trainf)])
#altera status para nome do arquivo
contexto = self.statusFile.get_context_id("fileAbrir")
self.statusFile.push(contexto,str(dialog.get_filename()))
except:
self.feedStatus.gerarStatus(self.feedStatus.contexto_load)
#pega a excecao gerada
trace = traceback.format_exc()
#imprime
#print "Ocorreu um erro: \n",trace
#salva em arquivo
file("trace.log","a").write(trace)
finally:
dialog.destroy()
def testFANN(paths, network):
net = nl.load(network)
predict = net.sim(np.array(getImages(paths))).tolist()
for n, x in enumerate(predict):
predict[n] = x[0]
real = getReal(paths)
return [real, predict]
def __init__(self, index):
threading.Thread.__init__(self)
self.index = index
self.c = Converter()
target = scipy.io.loadmat('./algorithms/t3_output.mat')
self.net = nl.load('./algorithms/t3_estimator.net')
t = target['t3_output']
self.norm_t = nl.tool.Norm(t)
def print_weights_of_saved_net(path):
try:
net = nl.load(path)
for layer in net.layers:
pprint(layer.np['w'])
except IOError as e:
logging.error("Can not open file to print weights of net")
logging.error(e)
def __init__(self, index):
threading.Thread.__init__(self)
self.index = index
self.c = Converter()
target = scipy.io.loadmat('./algorithms/t3_output.mat')
self.net = nl.load('./algorithms/t3_estimator.net')
t = target['t3_output']
self.norm_t = nl.tool.Norm(t)
def main(args, argv):
# PARSING ARGUMENTS <-- TEST_FOLDER, INPUT_NET, OUTPUT_FILE
test_input_data = list()
results = str()
if (args != 3):
exit(1)
try:
# Testing data
w_files_test = os.listdir(argv[0])
if not (argv[1].endswith("NET")):
exit(2)
w_file_output = open(argv[2], "w")
net = nl.load(argv[1])
except:
exit(3)
# LOADING TESTING DATA
print("\nLOADING TEST DATA\n")
w_files_test = wav_to_mfcc(argv[0], w_files_test, test_input_data)
# TESTING
print("\n\tTESTING with test data\n")
for index in range(len(test_input_data)):
score = net.sim([test_input_data[index]])[0][0]
file_name = w_files_test[index].split('/')
file_name = file_name[-1].split('.')
file_name = file_name[-2]
if (score > 0.92):
decision = 1
else:
decision = 0
if (score > 1):
score = 1
elif (score < 0):
score = 0
print(file_name + " " + str(score) + " " + str(decision))
results += file_name + " " + str(score) + " " + str(decision) + "\n"
w_file_output.write(results)
w_file_output.close()
print("\nPROCESSED\n")
return 0
def test_res(self, path_for_testing):
"""
Provide test of created ann for examples.
Parameters
----------
:param path_for_testing: str or unicode
Path to files for testing.
Returns
-------
:return:
Return statistic of tests.
"""
prc_sum, nm_sum = 0.0, 0.0
for fls in os.listdir(path_for_testing):
self.logger.info(u"Word %s" % fls)
nm_sum += 1
pron = extractor.MelExtractor(glob_path=path_for_testing+u"%s" % fls)
ext_res = pron.viewer()
prc_n, nm_n, prc_b, nm_b = 0.0, 0.0, 0.0, 0.0
for i in self.lst_of_commands:
try:
tmp = open(u'networks/%s_brain' % i)
nt = pickle.load(tmp)
tmp.close()
net = nl.load(u'networks/%s_neurolab' % i)
nm_n += 1
nm_b += 1
example = self.ext_t(ext_res)
if fls.startswith(i):
if round(net.sim([example])[0][0]) == 1.0:
prc_n += 1
if round(nt.activate(example)) == 1.0:
prc_b += 1
else:
if round(net.sim([example])[0][0]) == 0.0:
prc_n += 1
if round(nt.activate(example)) == 0.0:
prc_b += 1
except IOError:
self.logger.critical(u"no created networks for %s" % i)
if prc_n/nm_n*100 == 100.0:
self.logger.info(u'Word was recognized by Neurolab')
prc_sum += 1
else:
self.logger.debug(u'Neurolab %s' % unicode(prc_n/nm_n*100))
if prc_b/nm_b*100 == 100.0:
self.logger.info(u'Word was recognized by PyBrain')
else:
self.logger.debug(u'PyBrain %s' % unicode(prc_b/nm_b*100))
self.logger.info(u'Result is %s' % unicode(prc_sum/nm_sum*100))
print u"Result of recognition by Neurolab is %s percents" % unicode(prc_sum/nm_sum*100)
def test(data):
global net
net = nl.load('part2_16.txt')
count = 0
for i in data:
#print len(i[0]),type(i[0])
out = net.sim([i[0]])[0]
val = i[1]
#print out, val
if np.argmax(out) == val.index(max(val)):
count += 1
print "accuracy : ", (float(count) / len(data)) * 100
def test(data):
global net
net = nl.load('autoencoder2_4_second_weights_2.txt')
count = 0
for i in data:
#print len(i[0]),type(i[0])
out = i[0]
val = i[1]
#print out, val
if np.argmax(out) == val.index(max(val)):
count += 1
print "accuracy : ", (float(count) / len(data)) * 100
def run(from_file=False, to_file=False):
functions = [(f'damped_sine_wave_{i}', get_damped_sine_wave_fun(),
(0, 10 * math.pi, 10**i + 1)) for i in range(1, 3)]
functions.extend([
('rosenbrock_1', get_rosenbrock_fun(), (-2, 2, 11), (-1, 3, 11)),
# ('rosenbrock_2', get_rosenbrock_fun(), (-2, 2, 21), (-1, 3, 21)),
('3d_sine_1', get_3d_sine_fun(), (0, 10 * math.pi, 11),
(0, 10 * math.pi, 11)),
])
for f_name, f, *ranges in functions:
file_prefix = f'output/nl_{f_name}'
print(f"### Using neurolab package for training neural network to "
f"approximate {f_name.replace('_', ' ')} function on ranges: "
f"{ranges}. Output files prefix: {file_prefix}")
X, y = generate_fun_samples(f, *ranges)
X_scaler = preprocessing.StandardScaler()
X_scaled = X_scaler.fit_transform(X)
y_scaler = preprocessing.MaxAbsScaler()
y_scaled = y_scaler.fit_transform(y)
if from_file:
net = nl.load(file_prefix + NET)
with open(file_prefix + HIS, 'rb') as file:
training_history = pickle.load(file)
else:
layers = [NEURONS_BY_LAYER] * LAYERS_COUNT + [1]
net = nl.net.newff([[-2, 2] for _ in ranges], layers)
@timeit
def train():
return nl.train.train_gdm(net,
X_scaled,
y_scaled,
epochs=EPOCHS,
show=100)
training_history = train()
if to_file:
net.save(file_prefix + NET)
with open(file_prefix + HIS, 'wb') as file:
pickle.dump(training_history, file)
def simulate(X_sim):
return y_scaler.inverse_transform(
net.sim(X_scaler.transform(X_sim)))
plot_experiment_results(f, file_prefix, simulate, ranges, X, y,
training_history)
def some_func():
try:
net = nl.load(Conf.NET_FILE)
except IOError as e:
print e
songs_info_loader = SongsInfoLoader(Conf.SONG_INFO_WITH_ANNOTATIONS_PATH)
songs_list = songs_info_loader.get_songs_list()
# split song for development list and evaluate list
dev_songs_list, eval_song_list = split_songs_for_purpose(songs_list)
dev_input, dev_targets = change_to_proper_format(dev_songs_list)
eval_input, eval_targets = change_to_proper_format(eval_song_list)
result = net.sim(dev_targets)
pprint(result)
def get_activations(data, index):
net = nl.load('autoencoder2_16_second_weights.txt')
a = data[index]
# for i in range(len(net.layers)):
weights = []
biases = []
biases.append(net.layers[0].np['b'])
weights.append(net.layers[0].np['w'])
biases = np.array(biases)
weights = np.array(weights)
print "hello:", weights[0].shape, (np.transpose(np.dot(weights[0], a)) +
biases).shape
#print len(weights),len(weights[0]),len(weights[0][0]),weights[0].shape,np.transpose(a).shape,biases.shape
a = sigmoid_vec(np.add(np.transpose(np.dot(weights[0], a)), biases))
print a.shape
return a[0].tolist()
def test_neuro():
Ndata,Pdata = main()
p_data = []
x = len(Ndata)#N file number
y = len(Ndata[0])#channel number here is 8
o = len(Pdata)#P file number
p = len(Pdata[0])#channel number here is 8
bpnet=nl.load('relief_hurst.net')
answer = bpnet.sim(Ndata[7])
print answer
for i in range(o):
n_answer = bpnet.sim(Pdata[i])
print n_answer
def neuronas(self, filedir):
imagen = cv2.imread(filedir)
imagen = ecnontrar_tomate(imagen)
cv2.imwrite("tomate-recortado.jpg", imagen)
cadena = sacar_pixels("tomate-recortado.jpg")
if (os.path.exists("datos-tomate.csv") == True):
os.remove("datos-tomate.csv")
archivo_entrenamiento = open("datos-tomate.csv", "a")
archivo_entrenamiento.write(cadena)
archivo_entrenamiento.close()
datos = np.matrix(sp.genfromtxt("datos-tomate.csv", delimiter=" "))
rna = nl.load("red-neuronal-artificial.tmt")
salida = rna.sim(datos)
self.salida1(str(salida[0][0]))
self.salida2(str(salida[0][1]))
self.salida3(str(salida[0][2]))
podrido = salida[0][0] * 100
maduro = salida[0][1] * 100
verde = salida[0][2] * 100
if (podrido > 80.):
if (maduro > 40.):
resultado = "el tomate esta a punto de podrirse"
self.estado(resultado)
else:
resultado = "el tomate esta podrido"
self.estado(resultado)
elif (maduro > 80.):
if (podrido > 40.):
resultado = "el tomate esta pasandose de su madurez"
self.estado(resultado)
elif (verde > 40.):
resultado = "El tomate esta a punto de llegar a su madurez"
self.estado(resultado)
else:
resultado = "El tomate esta en su mejor punto"
self.estado(resultado)
elif (verde > 80.):
if (maduro > 40.):
resultado = "el tomate esta madurando"
self.estado(resultado)
else:
resultado = "el tomate esta verde"
self.estado(resultado)
def neuro_validate(x, y, timeseries, name):
import neurolab as nl
import numpy as np
size = len(x)
inp = x.reshape(size,3)
tar = y.reshape(size,1)
# data preprocessing
mean_tar = tar.mean(axis=0)[0]
std_tar = tar.std(axis=0)[0]
tar = (tar - mean_tar) / (2 * std_tar)
# add another feature
inp2 = np.zeros(size)
# add another feature
for i, k in enumerate(y):
inp2[i] = tar[i - 1][0]
inp2 = inp2.reshape(size,1)
inp = np.concatenate((inp,inp2), axis=1)
import os
cwd = os.getcwd()
sub_dir = cwd + '/training_models/neuro_network'
filename = 'model_' + name
#net = nl.load( os.path.join(sub_dir, filename) )
net = nl.load(filename)
# Simulate network
out = net.sim(inp).reshape(size)
out = out * 2 * std_tar + mean_tar
# Plot result
import pylab as pl
pl.subplot(111)
pl.plot(timeseries, y, '.-r', timeseries, out, 'p-b')
pl.legend(['train target', 'net output'])
pl.show()
def neuro_validate(x, y, timeseries, name):
import neurolab as nl
import numpy as np
size = len(x)
inp = x.reshape(size, 3)
tar = y.reshape(size, 1)
# data preprocessing
mean_tar = tar.mean(axis=0)[0]
std_tar = tar.std(axis=0)[0]
tar = (tar - mean_tar) / (2 * std_tar)
# add another feature
inp2 = np.zeros(size)
# add another feature
for i, k in enumerate(y):
inp2[i] = tar[i - 1][0]
inp2 = inp2.reshape(size, 1)
inp = np.concatenate((inp, inp2), axis=1)
import os
cwd = os.getcwd()
sub_dir = cwd + '/training_models/neuro_network'
filename = 'model_' + name
#net = nl.load( os.path.join(sub_dir, filename) )
net = nl.load(filename)
# Simulate network
out = net.sim(inp).reshape(size)
out = out * 2 * std_tar + mean_tar
# Plot result
import pylab as pl
pl.subplot(111)
pl.plot(timeseries, y, '.-r', timeseries, out, 'p-b')
pl.legend(['train target', 'net output'])
pl.show()
def testNL():
net = nl.load('test_tretia.net')
sct = mss()
while (True):
screen_np = np.array(sct.grab({'top': 0, 'left': 0, 'width': 600, 'height': 200}))
inputs = inputs_from_process_img(screen_np)
# # drawing of line ROI
# cv2.line(screen_np, (90, 155), (400, 155), (0, 0, 0), 1)
# cv2.line(screen_np, (90, 175), (400, 175), (0, 0, 0), 1)
# cv2.imshow("Screencapture:", screen_np)
# print("inputs: {}".format(inputs))
outputs = net.sim(inputs)
# print(inputs)
# print("outputs: {}".format(outputs))
th.start_new_thread(decide, (outputs,))
if cv2.waitKey(15) & 0xFF == ord('q'):
cv2.destroyAllWindows()
break
import Image
import json
import neurolab as nl
import matplotlib.pyplot as pl
entry=[]
with open("input.txt","r") as input_file:
entry=json.load(input_file)
target=[]
input=[]
for i in range(0,50):
if(i%5==0):
target.append(entry[1][i])
input.append(entry[0][i])
structure=[]
net=nl.load("train.data")
result=net.sim(input)
print result
#print euclidean_distance(target, result)
import neurolab as nl
import numpy as np
import svm
import functools
from multiprocessing import Pool, cpu_count
import random
def kernel(xm, xn):
return (1 + np.dot(xm, xn)) ** 8
random.seed()
train_num = 500
net = nl.load("../features/ann80.net")
print("Loaded NN")
all_data = np.loadtxt("../combinedData.txt")
#index_list = range(0, all_data.shape[0])
#train_list = random.sample(index_list, train_num)
#test_list = [x for x in index_list if x not in train_list]
target = all_data[:, 0]
pixel_info = all_data[:, 1:]
train_info = pixel_info[:train_num]
print train_info.shape
test_info = pixel_info[train_num:]
print("Loaded info")
#cuts the last layer off of ann making the net that makes the features
inputParams = [[-1, 1]] * pixel_info.shape[1]
cadena = sacar_pixels("tomate-recortado.jpg")
if(os.path.exists("datos-tomate.csv")== True):
os.remove("datos-tomate.csv")
archivo_entrenamiento = open("datos-tomate.csv", "a")
archivo_entrenamiento.write(cadena)
archivo_entrenamiento.close()
datos = np.matrix(sp.genfromtxt("datos-tomate.csv", delimiter=" "))
print datos.shape
rna = nl.load("red-neuronal-artificial.tmt")
salida = rna.sim(datos)
podrido = salida[0][0] * 100
maduro = salida[0][1] * 100
verde = salida[0][2] * 100
resultado = ""
if (podrido > 80.):
if (maduro > 40.):
resultado = "el tomate esta a punto de podrirse"
else:
resultado = "el tomate esta podrido"
elif (maduro > 80.):
with open("npq_training_dataset.p", "rb") as f:
inp, tar = pickle.load(f)
num_input_units = len(inp[0])
num_output_units = len(tar[0])
minmax = [[0, 1]] * num_input_units
# One of the thumb rule to set nh = 2/3 * (ni + no)
size = [(num_input_units + num_output_units) * 3 / 5, num_output_units]
inp = inp.reshape(len(inp), num_input_units)
tar = tar.reshape(len(tar), num_output_units)
trans = [nl.trans.TanSig()] * (len(size) - 1) + [nl.trans.LogSig()]
# Create network with n layers
#net = nl.net.newff(minmax, size, transf=trans)
net = nl.load(raw_input('Model name: '))
# Change traning func, by default uses train_bfgs
#net.trainf = nl.train.train_gdx # Gradient descent with momentum backpropagation and adaptive lr
# Change error func, by default uses SSE()
#net.errorf = nl.error.MSE()
goal = 0.01
print "*" * 50
print " #Samples: ", len(inp)
print " #Epochs: ", epochs
print " #Input Units: ", net.ci
print "#Hidden Units: ", size[0]
print "#Output Units: ", net.co
print " #Goal: ", goal
def fix_transf(filename, sz):
net = nl.load(filename)
new_net = net_from_layers([[-1.0, 1.0]] * 38, net.layers[:-1], [sz], last_lin=True)
new_net.save(filename)
# Create train samples
input = np.array(input.data)
target = np.asfarray(target.data)
input = input[: target.shape[0]]
# Create network with 2 layers and random initialized
#norm = Norm(input)
#input = norm(input)
print input.shape
print target.shape
print '----------',minmax(input)
net = nl.net.newff(minmax(input), [12, 4], transf = [nl.trans.TanSig(), nl.trans.LogSig()])
net.trainf = nl.train.train_bfgs
error = net.train(input, target, epochs=1000, show=10, goal=0.02)
print '-------',error[-1]
net.save('net.net')
#Simulate network
print '\nprinting the simulated output';
net=nl.load('net.net')
output = net.sim(input)
out = output
for i in range(len(output)):
m=max(output[i])
print '[',m,']',
for j in range(4):
if output[i,j] == m:
print j+1,
print ';;',target[i]
net.save('net.net')
#!/usr/bin/python
import neurolab as nl
import numpy as np
import input
import target
from neurolab.tool import minmax
# Create train samples
input = np.array(input.data)
target = np.asfarray(target.data)
input = input[:target.shape[0]]
net = nl.load('net.net')
output = net.sim(input)
err = 0
for i in range(len(output)):
m = max(output[i])
for j in range(4):
if output[i, j] == m:
if target[i][j] == 0:
err += 1
print len(target), err
# print i.shape
# print target.shape
# sys.exit()
norm_t = nl.tool.Norm(t)
t = norm_t(t)
# train and save the neural network
"""
net = nl.net.newff(nl.tool.minmax(i), [10, 1])
err = net.train(i, t, epochs=1000, show=1)
net.save('t3_estimator.net')
"""
# load the neural network for transcoding time estimation
net = nl.load("t3_estimator.net")
# test the input using the neural network
out = net.sim(i)
out = norm_t.renorm(out)
t = norm_t.renorm(t)
# print target['t'] - out
e = t - out
# print type(e)
# print e.shape
# print len(e)
j = 0
while j < len(e):
print t[j], " ", e[j]
def loadnet(self):
return nl.load(self.netconfig)
def loadnetwork(file): return nl.load(file)
def load(cls, agent):
dir = join(agent, cls.name)
comp = cls(dir)
comp.nn = nl.load(join(dir, "utility.net"))
return comp
for y in xrange(0, (15)):
for x in xrange(0, (15)):
datos[ubicacion] = matriz_imagen[x, y][0]
datos[ubicacion + 1] = matriz_imagen[x, y][1]
datos[ubicacion + 2] = matriz_imagen[x, y][2]
ubicacion = ubicacion + 3
datos = datos / 255
#a contendra los datos para pasar por la red neuronal
a = datos
#calidad es la red mejor entrenada, calidad2 le sigue y por ultimo calidad3
#redes para calidad: calidad.net, calidad2.net, calidad3.net.
#red4 es la mejor para madurez, le sigue red3, despues red y por ultimo red2
#redes para madurez: red4.net, red3.net, red.net, red2.net
#se carga la red neuronal a utilizar.
#no importa si es para calidad o para madurez, el programa detecta que es lo que se esta midiendo
net = nl.load("red4.net")
#Resultado1 es para comprobar si existe un error en la busqueda, y no encuentra nada (es como una bandera)
resultado1 = 0
#posicion en donde encuentra a cual se parece mas
posicion = 0
#simular en la red con los datos en a
out = net.sim([a])
#maximo es una variable auxiliar para ayudar a encontrar al que se parece mas
maximo = out[0][0]
for x in range(len(out[0])):
resultado1 = 1
#el siguiente if busca a cual se parece mas lo que acaba de encontrar
if out[0][x] > maximo:
maximo = out[0][x],
posicion = x
import pandas as pd
import numpy as np
import neurolab as nl
from neurolab import trans
# Read Data
I_valid = pd.read_csv('../train.csv')
X_labels = (I_valid.ix[:, 0].values).astype('int')
X_valid = (I_valid.ix[:, 1:].values).astype('int')
# Pre-processing
Y_valid = np.zeros((42000, 784))
for i in range(42000):
for j in range(784):
if (X_valid[i, j] > 200):
Y_valid[i, j] = 1
del I_valid, X_valid
# Test
net = nl.load('newff.net')
out = net.sim(Y_valid)
# Validate
j = 0
for i in range(42000):
if (np.argmax(out[i]) == X_labels[i]):
j = j + 1
print('Accuracy: ')
print(j / 42000)
training = open('nnTraining.txt', 'r').readlines()
trainingIN = []
trainingOUT = []
for line in training:
tokens = line.rstrip().split()
nextIn = [float(x) for x in tokens[3:]]
trainingIN.append(nextIn)
nextOut = [float(x) for x in tokens[:3]]
trainingOUT.append(nextOut)
#inputs = [ [-1.0, 1.0] for j in range( 0, 3 ) ]
#net = neurolab.net.newff( inputs, [10,18,3] )
#net.trainf = neurolab.train.train_rprop
#net.train( trainingIN, trainingOUT, show=10 )
#net = nl.load('final_network')# topology[10,3]
net = nl.load('final_network2') #topology[10,18,3]
Validation = open('nnvalidation.txt', 'r').readlines()
ValidationIN = []
ValidationOUT = []
for line in Validation:
tokens = line.rstrip().split()
nextIn = [float(x) for x in tokens[3:]]
ValidationIN.append(nextIn)
nextOut = [float(x) for x in tokens[:3]]
ValidationOUT.append(nextOut)
res = net.sim(ValidationIN)
j = 0
predicted = []
while j in range(0, len(res)):
predicted.append([1.0 if x == res[j].max() else -1.0 for x in res[j]])
j += 1
os.remove("dato-prueba.csv")
archivo = open("dato-prueba.csv", "a")
archivo.write(cadena)
archivo.close()
datos = np.matrix(sp.genfromtxt("dato-prueba.csv", delimiter=" "))
matiz = 0
i = 0
#while(i != 1200):
# matiz = matiz + datos[0][i]
# i = i + 3
#print (datos.shape)
rna = nl.load("red-entrenada.tmt")
salida = rna.sim(datos)
verde = salida[0][0] * 100
maduro = salida[0][1] * 100
podrido = salida[0][2] * 100
resultado = ""
if(gradoMadurez <= 30):
print("Su tiempo estimado de vida es: 7 días")
if(gradoMadurez >= 31 and gradoMadurez <= 60):
print("Su tiempo estimado de vida es: 4 días")
if(gradoMadurez >= 61 and gradoMadurez <= 85):
print("Su tiempo estimado de vida es : 2 días")
if(gradoMadurez >= 86 and gradoMadurez <= 100):
print("La naranja ya caducó")
import neurolab as nl, cPickle as pickle, sys, numpy as np, time
start_time = time.time()
def discretize(alist):
return np.array([0 if x < 0.5 else 1 for x in alist])
if len(sys.argv) != 1:
model = sys.argv[1]
else:
print "Command line argument missing! Input model's filename..."
sys.exit()
with open("keypair_data.p", "rb") as inptarfile:
data, target = pickle.load(inptarfile)
net = nl.load(model)
test = net.sim(data)
exp = target[0]
gen = discretize(test[0])
print "*" * 50
print " Expected: ", exp
print "Generated: ", gen
if (exp == gen).all():
print "\nSame!!!"
else:
print "\nNot same!!!"
print "*" * 50
import neurolab as nl
import numpy as np
import svm
import functools
from multiprocessing import Pool, cpu_count
import random
def kernel(xm, xn):
return (1 + np.dot(xm, xn))**8
random.seed()
train_num = 500
net = nl.load("../features/ann80.net")
print("Loaded NN")
all_data = np.loadtxt("../combinedData.txt")
#index_list = range(0, all_data.shape[0])
#train_list = random.sample(index_list, train_num)
#test_list = [x for x in index_list if x not in train_list]
target = all_data[:, 0]
pixel_info = all_data[:, 1:]
train_info = pixel_info[:train_num]
print train_info.shape
test_info = pixel_info[train_num:]
print("Loaded info")
#cuts the last layer off of ann making the net that makes the features
#print i.shape
#print target.shape
#sys.exit()
norm_t = nl.tool.Norm(t)
t = norm_t(t)
#train and save the neural network
'''
net = nl.net.newff(nl.tool.minmax(i), [10, 1])
err = net.train(i, t, epochs=1000, show=1)
net.save('t3_estimator.net')
'''
#load the neural network for transcoding time estimation
net = nl.load('t3_estimator.net')
#test the input using the neural network
out = net.sim(i)
out = norm_t.renorm(out)
t = norm_t.renorm(t)
#print target['t'] - out
e = t - out
#print type(e)
#print e.shape
#print len(e)
j = 0
while j < len(e):
print t[j], ' ', e[j]
training = open('nnTraining.txt', 'r').readlines()
trainingIN =[]
trainingOUT = []
for line in training:
tokens = line.rstrip().split()
nextIn = [float(x) for x in tokens[3:]]
trainingIN.append(nextIn)
nextOut = [float(x) for x in tokens[:3]]
trainingOUT.append(nextOut)
#inputs = [ [-1.0, 1.0] for j in range( 0, 3 ) ]
#net = neurolab.net.newff( inputs, [10,18,3] )
#net.trainf = neurolab.train.train_rprop
#net.train( trainingIN, trainingOUT, show=10 )
#net = nl.load('final_network')# topology[10,3]
net = nl.load('final_network2') #topology[10,18,3]
Validation = open('nnvalidation.txt', 'r').readlines()
ValidationIN =[]
ValidationOUT = []
for line in Validation:
tokens = line.rstrip().split()
nextIn = [float(x) for x in tokens[3:]]
ValidationIN.append(nextIn)
nextOut = [float(x) for x in tokens[:3]]
ValidationOUT.append(nextOut)
res = net.sim(ValidationIN)
j =0
predicted =[]
while j in range(0, len(res)):
predicted.append([1.0 if x == res[j].max() else -1.0 for x in res[j]])
j+=1
# In[7]:
#get the data in two arrays classifications and the actual data
#classificationTrain,trainData = getData("ZipDigits.train.txt")
#classificationTest,testData = getData("ZipDigits.test.txt")
#combined set since we will use or on seperation technique
classification,data = getData('combinedData.txt')
# In[8]:
# Create network with 256 inputs, 2 neurons in hidden layer
# And 256 in output layer
ann = nl.load('features/ann80.net')
size = 500
subData = data[:size]
#inputParams = [[-1, 1]] * len(subData[0])
#ann = nl.net.newff(inputParams, [80,256])
#ann.trainf = nl.train.train_rprop
# In[8]:
#cuts the last layer off of ann making the net that makes the features
inputParams = [[-1, 1]] * len(subData[0])
featureNet = nl.net.newff(inputParams, [80])
featureNet.layers[0].np['w'][:] = ann.layers[0].np['w']
featureNet.layers[0].np['b'][:] = ann.layers[0].np['b']
# -*- coding: utf-8 -*-
"""
Created on Fri Jul 1 21:01:18 2016
@author: meza
"""
import neurolab as nl
import numpy as np
net = nl.net.newff([[0,1]] * 3, [4,2])
net.save("test.net")
net = nl.load("test.net")
# show layer weights and biases
for i in range(0,len(net.layers)):
print "Net layer", i
print net.layers[i].np['w']
print "Net bias", i
print net.layers[i].np['b']
#try setting layer weights
net.layers[0].np['w'][:] = np.array ([[0,1,2],
[3,4,5],
[4,5,6],
[6,7,8]]
)
# show layer weights and biases
def main():
featureFile = open(
'C:/Users/Abhi/workspace/MalwareClassification/asmTestFeatures.csv',
'r')
inp = featureFile.readlines()
inp1 = []
for line in inp:
inp1.append(line.split(","))
#print inp1[10868]
inp1 = inp1[1:]
#print inp1[0]
labels = []
for i in range(len(inp1)):
labels.append(inp1[i][-1][:-1])
inp1[i] = inp1[i][0:-1]
print len(labels[0])
print len(inp1)
for i in range(len(inp1)):
for j in range(len(inp1[0])):
inp1[i][j] = float(inp1[i][j])
for i in range(len(inp1)):
nConstant = sum(inp1[i])
for j in range(len(inp1[0])):
inp1[i][j] = inp1[i][j] / nConstant
#print len(inp1)
#print out
inp2 = np.array(inp1)
#out2 = np.array(out)
#tar = out2.reshape(len(out2),1)
#print inp2
#print len(inp2[0])
#print tar
#print len(inp2)
#print len(tar)'''
'''okCount=0
notokCount=0
for item in inp2:
if sum(item)> 0.99:
okCount += 1
else:
notokCount +=1
print okCount
print notokCount'''
newnet = nl.load('classify4.net')
print "fine"
roundans = []
ans = newnet.sim(inp2)
for i in range(len(ans)):
if round(ans[i][0], 3) <= 0.15:
roundans.append(0.1)
elif round(ans[i][0], 3) <= 0.25:
roundans.append(0.2)
elif round(ans[i][0], 3) <= 0.35:
roundans.append(0.3)
elif round(ans[i][0], 3) <= 0.45:
roundans.append(0.4)
elif round(ans[i][0], 3) <= 0.55:
roundans.append(0.5)
elif round(ans[i][0], 3) <= 0.65:
roundans.append(0.6)
elif round(ans[i][0], 3) <= 0.75:
roundans.append(0.7)
elif round(ans[i][0], 3) <= 0.85:
roundans.append(0.8)
elif round(ans[i][0], 3) <= 0.95:
roundans.append(0.9)
else:
roundans.append(0.0)
roundans = np.array(roundans)
roundans = roundans.reshape(len(roundans), 1)
#print roundans
#calculateError(roundans,tar)'''
printprobs(labels, roundans)
def main():
featureFile = open('C:/Users/Abhi/workspace/MalwareClassification/asmTestFeatures.csv', 'r')
inp = featureFile.readlines()
inp1=[]
for line in inp:
inp1.append(line.split(","))
#print inp1[10868]
inp1 = inp1[1:]
#print inp1[0]
labels = []
for i in range(len(inp1)):
labels.append(inp1[i][-1][:-1])
inp1[i] = inp1[i][0:-1]
print len(labels[0])
print len(inp1)
for i in range(len(inp1)):
for j in range (len(inp1[0])):
inp1[i][j] = float(inp1[i][j])
for i in range(len(inp1)):
nConstant = sum(inp1[i])
for j in range (len(inp1[0])):
inp1[i][j] = inp1[i][j]/nConstant
#print len(inp1)
#print out
inp2 = np.array(inp1)
#out2 = np.array(out)
#tar = out2.reshape(len(out2),1)
#print inp2
#print len(inp2[0])
#print tar
#print len(inp2)
#print len(tar)'''
'''okCount=0
notokCount=0
for item in inp2:
if sum(item)> 0.99:
okCount += 1
else:
notokCount +=1
print okCount
print notokCount'''
newnet = nl.load('classify4.net')
print "fine"
roundans=[]
ans = newnet.sim(inp2)
for i in range(len(ans)):
if round(ans[i][0],3) <= 0.15:
roundans.append(0.1)
elif round(ans[i][0],3) <= 0.25:
roundans.append(0.2)
elif round(ans[i][0],3) <= 0.35:
roundans.append(0.3)
elif round(ans[i][0],3) <= 0.45:
roundans.append(0.4)
elif round(ans[i][0],3) <= 0.55:
roundans.append(0.5)
elif round(ans[i][0],3) <= 0.65:
roundans.append(0.6)
elif round(ans[i][0],3) <= 0.75:
roundans.append(0.7)
elif round(ans[i][0],3) <= 0.85:
roundans.append(0.8)
elif round(ans[i][0],3) <= 0.95:
roundans.append(0.9)
else:
roundans.append(0.0)
roundans = np.array(roundans)
roundans = roundans.reshape(len(roundans),1)
#print roundans
#calculateError(roundans,tar)'''
printprobs(labels,roundans)
def predictSingleImage(image, network="semhistograma.net"):
net = nl.load(network)
predict = net.sim(np.array([getValues(image)]))
return (round(predict[0][0]))
from matplotlib import pylab as pl
from pprint import pprint
def some_func():
try:
net = nl.load(Conf.NET_FILE)
except IOError as e:
print e
songs_info_loader = SongsInfoLoader(Conf.SONG_INFO_WITH_ANNOTATIONS_PATH)
songs_list = songs_info_loader.get_songs_list()
# split song for development list and evaluate list
dev_songs_list, eval_song_list = split_songs_for_purpose(songs_list)
dev_input, dev_targets = change_to_proper_format(dev_songs_list)
eval_input, eval_targets = change_to_proper_format(eval_song_list)
result = net.sim(dev_targets)
pprint(result)
if __name__ == "__main__":
err = open_pickle(Conf.PICKLED_ERR_PATH)
processed_songs = open_pickle(Conf.PICKLED_SONGS_PATH)
net = nl.load(Conf.NET_FILE)
# ev = net.sim(eval_data)
pl.plot(err)
pl.show()
def importW (self, fin):
if check.isfile (fin):
self.nn = nl.load (fin)
def reg_net(self, reg_data):
net = nl.load("test.net")
inp = np.array(reg_data)
out = net.sim(inp)
return out
import matplotlib
import matplotlib.pyplot as plt
#img=matplotlib.image.imread('./test.png')
#print img
import neurolab as nl
net = nl.load('autoencoder2_16_second_weights.txt')
img = list(map(lambda x: x * 255, net.layers[0].np['w']))
img1 = [img[i:i + 2] for i in range(len(img) - 2)]
img2 = [img1[i:i + 16] for i in range(len(img1) - 16)]
print img
plt.imshow(img, cmap='gray', vmin=0, vmax=255) #,origin='lower')
plt.show()
with open("m2c_rsa_generate.p", "rb") as f:
inp, tar = pickle.load(f)
num_input_units = len(inp[0])
num_output_units = len(tar[0])
minmax = [[0, 1]] * num_input_units
size = [3 * num_input_units * 3 / 5, 2 * num_input_units, num_output_units]
inp = inp.reshape(len(inp), num_input_units)
tar = tar.reshape(len(tar), num_output_units)
trans = [nl.trans.TanSig()] * (len(size) - 1) + [nl.trans.LogSig()]
# Create network with n layers
#net = nl.net.newff(minmax, size, transf=trans)
net = nl.load(raw_input('Model name: '))
# Change traning func, by default uses train_bfgs
#net.trainf = nl.train.train_gdx # Gradient descent with momentum backpropagation and adaptive lr
npq_net = nl.load(raw_input('NPQ Model name: '))
net.layers[0].np['w'][:] = npq_net.layers[0].np['w'][:]
net.layers[0].np['b'][:] = npq_net.layers[0].np['b'][:]
net.layers[1].np['w'][:] = npq_net.layers[1].np['w'][:]
net.layers[1].np['b'][:] = npq_net.layers[1].np['b'][:]
goal = 0.01
print "*" * 50
print "#Training Samples: ", len(inp)