def evaluate_wass_dist(dataset_name, fake_samples):
gan_algorithm = 'WGAN'
opts = load_opts(dataset_name, gan_algorithm)
# update work dir
#opts['work_dir'] = '../' + opts['work_dir']
print(opts['work_dir'])
if not os.path.isdir(opts['work_dir']):
print('No working directory')
data = datahandler.DataHandler(opts)
data._load_data(opts)
# load fake samples
#fake_samples = np.load(opts['work_dir'] + '/samples.npy')
#print(fake_samples.shape)
opts['fake_points'] = fake_samples
g = gan_test.WassersteinGAN(opts, data)
g.evaluate()
def main():
dataset_name = sys.argv[1]
gan_algorithm = sys.argv[2]
# loads options relevant to the dataset, gan algorithm
opts = load_opts(dataset_name, gan_algorithm)
if os.path.isdir(opts['work_dir']):
shutil.rmtree(opts['work_dir'])
os.mkdir(opts['work_dir'])
# loads train, test datasets
data = datahandler.DataHandler(opts)
data._load_data(opts)
gan_dict = {
'AdaGAN': gan.AdaGAN,
'UnrolledGAN': gan.UnrolledGAN,
'WGAN': gan.WassersteinGAN,
'VEEGAN': gan.VEEGAN
}
# closes the tf graph
with gan_dict[gan_algorithm](opts, data) as g:
# trains the GAN
g.train()
# sample
samples = g._sample_internal()
# save samples and loss plots
np.save(os.path.join(opts['work_dir'], 'samples'), samples)
np.save(os.path.join(opts['work_dir'], 'epoch_g_loss'),
g._epoch_g_loss)
np.save(os.path.join(opts['work_dir'], 'epoch_d_loss'),
g._epoch_d_loss)
# compute some metrics
"""
# USER PROGRAMMER
import sys
import datahandler
if not len(sys.argv) == 3 and not len(sys.argv) == 4:
print("usage : python main.py <exel filename> <total_avrg> <sd = 20>")
exit(-1) # exit()는 종료하라는 의미이며, -1을 쓴 것은 비정상적인 종료라는 것을 알려주는 것
dh = datahandler.DataHandler(sys.argv[1])
if len(sys.argv) == 3:
dh.get_evaluation(sys.argv[2])
elif len(sys.argv) == 4:
dh.get_evaluation(int(sys.argv[2]), int(sys.argv[3]))
from __future__ import print_function
import datahandler
import networkhandler
import rootmodel
import keras
from keras.datasets import cifar10
from keras.layers import concatenate
#import cifar
batch_size = 32
epochs = 200
lrate = 0.01
decay = lrate / epochs
#init and load cifar dataset
bin_data = datahandler.DataHandler()
bin_data.load_cifar_data_set(2, "binary")
bin_data.normalize(255)
#dataset.sort_data_by_label(dataset.x_train, dataset.y_train)
#init network
root_network = networkhandler.Network("binary", batch_size, 2, epochs, True,
bin_data)
#define root network model
root_network.define_model(rootmodel.newModel(bin_data.x_train, 2))
root_network.preprocess()
opt = keras.optimizers.SGD(lr=lrate, momentum=0.9, decay=decay, nesterov=False)
root_network.compile(opt)
print(root_network.model.summary())
def __init__(self, filename=None):
self.dataHandler = datahandler.DataHandler(filename)
def __init__(self):
self.datahandler = datahandler.DataHandler()
self.foodOptions = self.datahandler.getData()
def data_handler():
dh = datahandler.DataHandler()
return dh
import datahandler
# 학년 전체 평균 : 50
dh = datahandler.DataHandler('class_2-3.xlsx')
dh.get_evaluation(50)
LAG_DAYS = 3 startdate = '20000101' # YYYYMMDD indices = ["%5EGSPC", "%5EIXIC", "%5EFVX", "%5ETYX", "%5EXMI", "%5ENYA"] #Neural Network INPUT = len(indices) * (LAG_DAYS+1) HIDDEN = 12 OUTPUT = 1 #Training ITERATIONS = 20 LRATE = 0.4 MOMENTUM = 0.6 data = dh.DataHandler() data.load_indices(indices, startdate, LAG_DAYS) data.create_data(INPUT, OUTPUT) train, test = data.get_datasets(TRAINING_PERCENT) print "Training:", len(train), "Testing:", len(test) sp_net = nh.NetHandler(INPUT, HIDDEN, OUTPUT, data) train_errors, val_errors = sp_net.train(train, LRATE, MOMENTUM, ITERATIONS) out_ser = sp_net.get_output(test, TRAINING_PERCENT) print "Net Topology: %d-%d-%d" % (INPUT, HIDDEN, OUTPUT) print sp_net.change_tomorrow() correct = 0 total = 0 misses = 0