def cpu_single(P, num_of_hidden, pop_size=1):
# number of input covariates
num_of_inputs = len(P[0])
# Tanh to keep implementation details easy in opencl
net = build_feedforward(num_of_inputs, num_of_hidden, 1, output_function="tanh")
@benchmark
def many_sim():
for n in xrange(pop_size):
net.sim(P)
many_sim()
def find_solution(P, T):
#test, validation = get_validation_set(P, T, validation_size = 0.33)
net = build_feedforward(input_number = len(P[0]), hidden_number = 4, output_number = len(T[0]))
#com = build_feedforward_committee(size = 4, input_number = len(P[0]), hidden_number = 6, output_number = len(T[0]))
epochs = 1000
testset, valset = get_validation_set(P, T, validation_size = 0.01)
print("Training...")
net = benchmark(train_evolutionary)(net, testset, valset, 100, random_range = 1)
net = benchmark(traingd)(net, testset, valset, epochs, learning_rate = 0.1, block_size = 1)
#benchmark(train_committee)(com, train_evolutionary, P, T, 100, random_range = 1)
#benchmark(train_committee)(com, traingd, P, T, epochs, learning_rate = 0.1, block_size = 30)
#P, T = test
Y = net.sim(P)
area, best_cut = plotroc(Y, T, 1)
plot2d2c(net, P, T, figure = 2, cut = best_cut)
#P, T = validation
#Y = com.sim(P)
#plotroc(Y, T, 2)
# print("")
# print("Stats for cut = 0.5")
# [num_correct_first, num_correct_second, total_performance, num_first, num_second, missed] = stat(Y, T)
#save_network(best, "/export/home/jonask/Projects/aNeuralN/ANNs/classification_gdblock20_rocarea" + str(area) + ".ann")
#save_network(best, "/export/home/jonask/Projects/aNeuralN/ANNs/classification_genetic_rocarea" + str(area) + ".ann")
#save_committee(com, "/export/home/jonask/Projects/aNeuralN/ANNs/classification_gdblock30_rocarea" + str(area) + ".anncom")
#save_committee(com, "/export/home/jonask/Projects/aNeuralN/ANNs/classification_genetic_rocarea" + str(area) + ".anncom")
plt.show()
Created on Jun 7, 2011
@author: jonask
'''
from kalderstam.neural.error_functions.sum_squares import total_error
from kalderstam.neural.network import build_feedforward
from kalderstam.util.filehandling import parse_data
from kalderstam.neural.training.gradientdescent import traingd
from kalderstam.neural.training.davis_genetic import train_evolutionary
import numpy
xor_set = [[0, 0, 0],
[0, 1, 1],
[1, 0, 1],
[1, 1, 0]]
xor_set = numpy.array(xor_set)
P, T = parse_data(xor_set, targetcols = 2, inputcols = [0, 1], normalize = False)
net = build_feedforward(2, 4, 1)
print("Error before training: " + str(total_error(T, net.sim(P))))
net = traingd(net, (P, T), (None, None), epochs = 1000, learning_rate = 0.1, block_size = 0)
print("Error after training: " + str(total_error(T, net.sim(P))))
net = build_feedforward(2, 4, 1)
print("Error before genetic training: " + str(total_error(T, net.sim(P))))
net = train_evolutionary(net, (P, T), (None, None), epochs = 100, population_size = 100)
print("Error after genetic training: " + str(total_error(T, net.sim(P))))
print('Aaawww....')
outputs = net.sim(P)
plot_network_weights(net)
plt.figure()
plt.title('Scatter plot sum square error\n' + filename)
plt.xlabel('Survival time years')
plt.ylabel('Network output')
try:
plt.scatter(T.flatten(), outputs.flatten(), c = 'g', marker = 's')
plt.plot(T.flatten(), T.flatten(), 'r-')
except:
pass
if __name__ == "__main__":
logging.basicConfig(level = logging.INFO)
glogger.setLoggingLevel(glogger.info)
p = 4 #number of input covariates
#net = load_network('/home/gibson/jonask/Projects/aNeuralN/ANNs/PERCEPTRON.ann')
net = build_feedforward(p, 10, 1, output_function = "linear")
lineartarget_nn = '/home/gibson/jonask/Dropbox/Ann-Survival-Phd/fake_data_set/lineartarget_no_noise.txt'
nonlineartarget_nn = '/home/gibson/jonask/Dropbox/Ann-Survival-Phd/fake_data_set/nonlineartarget_no_noise.txt'
lineartarget_wn = '/home/gibson/jonask/Dropbox/Ann-Survival-Phd/fake_data_set/lineartarget_with_noise.txt'
nonlineartarget_wn = '/home/gibson/jonask/Dropbox/Ann-Survival-Phd/fake_data_set/nonlineartarget_with_noise.txt'
while True:
experiment(net, nonlineartarget_wn, 500)
plt.show()
filename = "/home/gibson/jonask/Dropbox/Ann-Survival-Phd/Two_thirds_of_SA_1889_dataset.txt"
P, T = parse_file(filename, targetcols = [4, 5], inputcols = [2, -4, -3, -2, -1], ignorerows = [0], normalize = True)
#P, T = parse_file(filename, targetcols = [4, 5], inputcols = [2, -3], ignorerows = [0], normalize = True)
#Remove tail censored
P, T = copy_without_tailcensored(P, T)
#Limit to incourage overtraining!
#rows = sample(range(len(T)), 100)
#P = P[rows]
#T = T[rows]
p = len(P[0]) #number of input covariates
#net = load_network('/home/gibson/jonask/Projects/aNeuralN/ANNs/4x10x10x1.ann')
net = build_feedforward(p, 30, 1, output_function = 'linear')
#Initial state
outputs = net.sim(P)
orderscatter(outputs, T, filename, 's')
for var in xrange(len(P[0, :])):
plot_input(P[:, var])
glogger.show()
epochs = 20000
rate = 1
block_size = 0
for times in range(100):
kwargs['one'] = 1
queue.put((car(), args, kwargs))
w.join()
#Now try with a neural network
P1, T1 = loadsyn1(100000)
P2, T2 = loadsyn1(100000)
P3, T3 = loadsyn1(100000)
P4, T4 = loadsyn1(100000)
P5, T5 = loadsyn1(100000)
P6, T6 = loadsyn1(100000)
P7, T7 = loadsyn1(100000)
P8, T8 = loadsyn1(100000)
net = build_feedforward(2, 1, 1)
nw1 = Neural_Worker(queue)
nw1.start()
nw2 = Neural_Worker(queue)
nw2.start()
queue.put((net, [P1], {}))
queue.put((net, [P2], {}))
nw1.join()
nw2.join()
print "Testing pool!"
p = Pool()
def train_single():
try:
netsize = input('Number of hidden nodes? [3]: ')
except SyntaxError as e:
netsize = 3
try:
pop_size = input('Population size? [50]: ')
except SyntaxError as e:
pop_size = 50
try:
mutation_rate = input('Please input a mutation rate (0.25): ')
except SyntaxError as e:
mutation_rate = 0.25
SB22 = "/home/gibson/jonask/Dropbox/Ann-Survival-Phd/Two_thirds_of_SA_1889_dataset_SB22.txt"
Benmargskohorten = "/home/gibson/jonask/Dropbox/Ann-Survival-Phd/Two_thirds_of_SA_1889_dataset_Benmargskohorten.txt"
SB91b = "/home/gibson/jonask/Dropbox/Ann-Survival-Phd/Two_thirds_of_SA_1889_dataset_SB91b.txt"
all_studies = "/home/gibson/jonask/Dropbox/Ann-Survival-Phd/Two_thirds_of_SA_1889_dataset.txt"
#Real data
print("Studies to choose from:")
print("1: SB22")
print("2: Benmargskohorten")
print("3: SB91b")
print("0: All combined (default)")
try:
study = input("Which study to train on? [0]: ")
except SyntaxError as e:
study = 0
if study == 1:
filename = SB22
elif study == 2:
filename = Benmargskohorten
elif study == 3:
filename = SB91b
else:
filename = all_studies
try:
columns = input("Which columns to include? (Do NOT forget trailing comma if only one column is used, e.g. '3,'\nAvailable columns are: 2, -4, -3, -2, -1. Just press ENTER for all columns.\n")
except SyntaxError:
columns = (2, -4, -3, -2, -1)
#P, T = parse_file(filename, targetcols = [4, 5], inputcols = [2, -4, -3, -2, -1], ignorerows = [0], normalize = True)
P, T = parse_file(filename, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
#Used for output comparison
studies = {}
studies[SB22] = parse_file(SB22, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
studies[Benmargskohorten] = parse_file(Benmargskohorten, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
studies[SB91b] = parse_file(SB91b, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
studies[all_studies] = parse_file(all_studies, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
#remove tail censored
#P, T = copy_without_tailcensored(P, T)
#Divide into validation sets
#((tP, tT), (vP, vT)) = get_validation_set(P, T, validation_size = 0.25, binary_column = 1)
TandV = get_cross_validation_sets(P, T, 2 , binary_column = 1)
#Network part
p = len(P[0]) #number of input covariates
net = build_feedforward(p, netsize, 1, output_function = 'linear')
#net = build_feedforward_multilayered(p, [7, 10], 1, output_function = 'linear')
try:
epochs = input("Number of generations (200): ")
except SyntaxError as e:
epochs = 200
for times, ((tP, tT), (vP, vT)) in zip(xrange(2), TandV):
#train
net = test(net, tP, tT, vP, vT, filename, epochs, population_size = pop_size, mutation_rate = mutation_rate)
raw_input("Press enter to show plots...")
glogger.show()
def train_single():
try:
netsize = input('Number of hidden nodes? [1]: ')
except SyntaxError as e:
netsize = 1
try:
pop_size = input('Population size? [100]: ')
except SyntaxError as e:
pop_size = 100
try:
mutation_rate = input('Please input a mutation rate (0.05): ')
except SyntaxError as e:
mutation_rate = 0.05
SB22 = "/home/gibson/jonask/Dropbox/Ann-Survival-Phd/Two_thirds_of_SA_1889_dataset_SB22.txt"
Benmargskohorten = "/home/gibson/jonask/Dropbox/Ann-Survival-Phd/Two_thirds_of_SA_1889_dataset_Benmargskohorten.txt"
SB91b = "/home/gibson/jonask/Dropbox/Ann-Survival-Phd/Two_thirds_of_SA_1889_dataset_SB91b.txt"
all_studies = "/home/gibson/jonask/Dropbox/Ann-Survival-Phd/Two_thirds_of_SA_1889_dataset.txt"
#Real data
print("Studies to choose from:")
print("1: SB22")
print("2: Benmargskohorten")
print("3: SB91b")
print("0: All combined (default)")
try:
study = input("Which study to train on? [0]: ")
except SyntaxError as e:
study = 0
if study == 1:
filename = SB22
elif study == 2:
filename = Benmargskohorten
elif study == 3:
filename = SB91b
else:
filename = all_studies
try:
columns = input("Which columns to include? (Do NOT forget trailing comma if only one column is used, e.g. '3,'\nAvailable columns are: 2, -4, -3, -2, -1. Just press ENTER for all columns.\n")
except SyntaxError:
columns = (2, -4, -3, -2, -1)
#P, T = parse_file(filename, targetcols = [4, 5], inputcols = [2, -4, -3, -2, -1], ignorerows = [0], normalize = True)
P, T = parse_file(filename, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
#Used for output comparison
studies = {}
studies[SB22] = parse_file(SB22, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
studies[Benmargskohorten] = parse_file(Benmargskohorten, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
studies[SB91b] = parse_file(SB91b, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
studies[all_studies] = parse_file(all_studies, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
#remove tail censored
try:
cutoff = input('Cutoff for censored data? [9999 years]: ')
except SyntaxError as e:
cutoff = 9999
P, T = copy_without_censored(P, T, cutoff)
#Divide into validation sets
try:
pieces = input('Size of validation set? Input denominator (1 for no validation set). Default is 1/[1] parts: ')
except:
pieces = 1
TandV = get_cross_validation_sets(P, T, pieces , binary_column = 1)
#Network part
p = len(P[0]) #number of input covariates
net = build_feedforward(p, netsize, 1, output_function = 'linear')
#net = build_feedforward_multilayered(p, [7, 10], 1, output_function = 'linear')
#Initial state
#outputs = net.sim(tP)
#orderscatter(outputs, tT, filename, 's')
try:
epochs = input("Number of generations (1): ")
except SyntaxError as e:
epochs = 1
for ((tP, tT), (vP, vT)) in TandV:
#train
net = test(net, tP, tT, vP, vT, filename, epochs, population_size = pop_size, mutation_rate = mutation_rate)
if plt:
outputs = net.sim(tP)
threshold = kaplanmeier(time_array = tT[:, 0], event_array = tT[:, 1], output_array = outputs[:, 0])
if len(vP) > 0:
outputs = net.sim(vP)
kaplanmeier(time_array = vT[:, 0], event_array = vT[:, 1], output_array = outputs[:, 0], threshold = threshold)
print("\nThreshold dividing the training set in two equal pieces: " + str(threshold))
raw_input("\nPress enter to show plots...")
plt.show()
try:
answer = input("Do you wish to print network output? Enter filename, or 'no' / 'n'. ['n']: ")
except (SyntaxError, NameError):
answer = 'n'
if os.path.exists(answer):
print("File exists. Will add random number to front")
answer = str(random.randint(0, 123456)) + answer
if answer != 'n' and answer != 'no':
print_output(answer, net, filename, targetcols = [4, 5], inputcols = columns, ignorerows = [0], normalize = True)
