def main():
# sample_sizes = [int(sys.argv[1]), int(sys.argv[2])]
sample_0 = 20
sample_1 = 200
values = 80
seed = int(sys.argv[1])
tot_ = 150
# seed = 4
np.random.seed(seed)
results = {}
tf.reset_default_graph()
sess = tf.InteractiveSession()
#create sin_data
data = json.loads(open('data_light_curves_2.json').read())
curve_0 = np.asarray(data['2'])
curve_0_x = curve_0
curve_0_y = np.asarray([np.arange(values)] * sample_0)
# curve_0_x, curve_0_y =
#sample_curves(curve_0, values) #random sampling curves
#create saw_data
curve_1 = np.asarray(data['3'])
curve_1_x = curve_1
curve_1_y = np.asarray([np.arange(values)] * sample_1)
#curve_1_x, curve_1_y =
#sample_curves(curve_1, values)
print(curve_0_y[:sample_0].shape, curve_0_x[:sample_0].shape)
#GAN first curve
curve_0_y_test = np.asarray([np.arange(values)] * (tot_ - sample_0))
t_0_g_out, curves_0_g_out = routine(sess, curve_0_x[:sample_0],
curve_0_y[:sample_0],
curve_0_x[sample_0:tot_ - sample_0],
curve_0_y_test, 10, values, sample_0,
tot_ - sample_0, False)
print('first')
tf.reset_default_graph()
sess = tf.InteractiveSession()
#GAN second curve
curve_1_y_test = np.asarray([np.arange(values)] * (tot_ - sample_1))
t_1_g_out, curves_1_g_out = routine(sess, curve_1_x[:sample_1],
curve_1_y[:sample_1],
curve_1_x[sample_1:tot_ - sample_1],
curve_1_y_test, 10, values, sample_1,
tot_ - sample_1, False)
# CLASSIFIER ON REAL VALUES
train_data_classifier_curve_0 = np.concatenate(
(curve_0_x[:sample_0], curve_0_y[:sample_0]), axis=1)
train_data_classifier_curve_1 = np.concatenate(
(curve_1_x[:sample_1], curve_1_y[:sample_1]), axis=1)
train_data_classifier = np.concatenate(
(train_data_classifier_curve_0, train_data_classifier_curve_1), axis=0)
classificatore_real = classifier(values)
cl_real = classificatore_real.create_clas()
num_rounds = 600
label_x = [[int(i)] for i in np.zeros(sample_0)]
label_y = [[int(i)] for i in np.ones(sample_1)]
labels = np.concatenate((label_x, label_y), axis=0)
classificatore_real.train(train_data_classifier, labels, cl_real,
num_rounds)
#CLASSIFIER ON GAN VALUES
train_x_0 = np.concatenate((curve_0_x[:sample_0], curves_0_g_out), axis=0)
train_y_0 = np.concatenate((curve_0_y[:sample_0], t_0_g_out), axis=0)
train_x_1 = curve_1_x[:sample_1]
# train_x_1 = np.concatenate((curve_1_x[:sample_1], curves_1_g_out), axis=0)
train_y_1 = curve_1_y[:sample_1]
#train_y_1 = np.concatenate((curve_1_y[:sample_1], t_1_g_out), axis=0)
train_data_classifier_curve_0 = np.concatenate(
(train_x_0[:tot_], train_y_0[:tot_]), axis=1)
train_data_classifier_curve_1 = np.concatenate(
(train_x_1[:tot_], train_y_1[:tot_]), axis=1)
train_data_classifier = np.concatenate(
(train_data_classifier_curve_0, train_data_classifier_curve_1), axis=0)
classificatore = classifier(values)
cl = classificatore.create_clas()
num_rounds = 600
label_x = [[int(i)] for i in np.zeros(tot_)]
label_y = [[int(i)] for i in np.ones(tot_)]
labels = np.concatenate((label_x, label_y), axis=0)
classificatore.train(train_data_classifier, labels, cl, num_rounds)
curve_0_y_ = np.asarray([np.arange(values)] * 100)
test_data_classifier_curve_0 = np.concatenate(
(curve_0_x[sample_0:sample_0 + 100], curve_0_y_), axis=1)
test_data_classifier_curve_1 = np.concatenate(
(curve_1_x[sample_1:sample_1 + 100], curve_0_y_), axis=1)
test_data = np.concatenate(
(test_data_classifier_curve_0, test_data_classifier_curve_1), axis=0)
label_x_test = [[int(i)] for i in np.zeros(100)]
label_y_test = [[int(i)] for i in np.ones(100)]
labels_test = np.concatenate((label_x_test, label_y_test), axis=0)
#evaluate classifier on real curve
acc_class_real = classificatore.valuate(test_data, labels_test,
cl)['accuracy']
acc_class_real_real = classificatore_real.valuate(test_data, labels_test,
cl_real)['accuracy']
results['accuracy_real'] = float(acc_class_real)
results['accuracy_real_real'] = float(acc_class_real_real)
# results['loss_sin'] = [g_loss_sin, d_loss_sin]
# results['loss_saw'] = [g_loss_saw, d_loss_saw]
out_file_ac = 'results.s{}.sam{}.noise{}.json'.format(
seed, sample_0, values)
open(out_file_ac, 'w').write(json.dumps(results))
out_file_curves = 'curves.s{}.sam{}.noise{}.json'.format(
seed, sample_0, values)
open(out_file_curves, 'wb').write(
pickle.dumps([[curves_0_g_out, t_0_g_out], [curves_1_g_out,
t_1_g_out]]))
return results
def main():
# sample_sizes = [int(sys.argv[1]), int(sys.argv[2])]
sample_0 = 100
sample_1 = 200
values = 24
seed = int(sys.argv[1])
tot_ = 300
size = 8
num = int(values / size)
# seed = 4
np.random.seed(seed)
results = {}
tf.reset_default_graph()
sess = tf.InteractiveSession()
#create sin_data
data = json.loads(open('PowerDemand.json').read())
curve_0 = np.asarray(data['1'])
curve_0_x = curve_0
curve_0_y = np.asarray([np.arange(values)] * sample_0)
# curve_0_x, curve_0_y =
#sample_curves(curve_0, values) #random sampling curves
#create saw_data
curve_1 = np.asarray(data['2'])
curve_1_x = curve_1
curve_1_y = np.asarray([np.arange(values)] * sample_1)
#curve_1_x, curve_1_y =
#sample_curves(curve_1, values)
slice_x_0, slice_y_0 = create_slices(curve_0_x, size)
slice_x_1, slice_y_1 = create_slices(curve_1_x, size)
train_x_0 = slice_x_0[:num * sample_0]
train_y_0 = slice_y_0[:num * sample_0]
train_x_1 = slice_x_1[:num * sample_1]
train_y_1 = slice_y_1[:num * sample_1]
#CLASSIFIER ON GAN VALUES
train_x_0 = np.concatenate((curve_0_x[:sample_0], curves_0_g_out), axis=0)
train_y_0 = np.concatenate((curve_0_y[:sample_0], t_0_g_out), axis=0)
#train_x_1 = curve_1_x[:sample_1]
print(np.asarray(curve_1_x).shape, np.asarray(curves_1_g_out).shape)
train_x_1 = np.concatenate((curve_1_x[:sample_1], curves_1_g_out), axis=0)
#train_y_1 = curve_1_y[:sample_1]
train_y_1 = np.concatenate((curve_1_y[:sample_1], t_1_g_out), axis=0)
train_data_classifier_curve_0 = np.concatenate(
(train_x_0[:tot_], train_y_0[:tot_]), axis=1)
train_data_classifier_curve_1 = np.concatenate(
(train_x_1[:tot_], train_y_1[:tot_]), axis=1)
train_data_classifier = np.concatenate(
(train_data_classifier_curve_0, train_data_classifier_curve_1), axis=0)
classificatore = classifier(values)
cl = classificatore.create_clas()
num_rounds = 900
label_x = [[int(i)]
for i in np.zeros(train_data_classifier_curve_0.shape[0])]
label_y = [[int(i)]
for i in np.ones(train_data_classifier_curve_1.shape[0])]
labels = np.concatenate((label_x, label_y), axis=0)
print(labels.shape, train_data_classifier.shape)
classificatore.train(train_data_classifier, labels, cl, num_rounds)
curve_0_y_ = np.asarray([np.arange(values)] * 200)
test_data_classifier_curve_0 = np.concatenate(
(curve_0_x[sample_0:sample_0 + 200], curve_0_y_), axis=1)
test_data_classifier_curve_1 = np.concatenate(
(curve_1_x[sample_1:sample_1 + 200], curve_0_y_), axis=1)
test_data = np.concatenate(
(test_data_classifier_curve_0, test_data_classifier_curve_1), axis=0)
label_x_test = [[int(i)] for i in np.zeros(200)]
label_y_test = [[int(i)] for i in np.ones(200)]
labels_test = np.concatenate((label_x_test, label_y_test), axis=0)
#evaluate classifier on real curve
acc_class_real = classificatore.valuate(test_data, labels_test,
cl)['accuracy']
acc_class_real_real = classificatore_real.valuate(test_data, labels_test,
cl_real)['accuracy']
results['accuracy_real'] = float(acc_class_real)
results['accuracy_real_real'] = float(acc_class_real_real)
# results['loss_sin'] = [g_loss_sin, d_loss_sin]
# results['loss_saw'] = [g_loss_saw, d_loss_saw]
out_file_ac = 'results.s{}.sam{}.noise{}.json'.format(
seed, sample_0, values)
open(out_file_ac, 'w').write(json.dumps(results))
out_file_curves = 'curves.s{}.sam{}.noise{}.json'.format(
seed, sample_0, values)
open(out_file_curves, 'wb').write(
pickle.dumps([[curves_0_g_out, t_0_g_out], [curves_1_g_out,
t_1_g_out]]))
return results
def main():
# sample_sizes = [20]
sample_sizes = [int(sys.argv[1])]
sample = sample_sizes[0]
# noise_sizes = [15]
noise_sizes = [int(sys.argv[2])]
values = noise_sizes[0]
seed = int(sys.argv[3])
# seed = 4
np.random.seed(seed)
results = {}
len_t = sample
tf.reset_default_graph()
sess = tf.InteractiveSession()
#create sin_data
sin_data = sample_data(
'sin', n_samples=sample*4, no_drop=values, reuse=False, mul_range=[1, 2])
sin_data.create_points()
#create saw_data
saw_data = sample_data('saw', n_samples=sample*4, mul_range=[
1, 2], no_drop=values, pred=1)
saw_data.create_points()
#GAN sin
Gz_sin, d_g_sin, d_x_sin, g_loss_sin, d_loss_sin, y_out_sin, curves_g_sin = routine(
sess, sin_data.y_values[:len_t], sin_data.x_values[:len_t],sin_data.y_values[len_t:], sin_data.x_values[len_t:], 10, values, len_t, len_t*3, False)
tf.reset_default_graph()
sess = tf.InteractiveSession()
#GAN saw
Gz_saw, d_g_saw, d_x_saw, g_loss_saw, d_loss_saw, y_out_saw, curves_g_saw = routine(
sess, saw_data.y_values[:len_t], saw_data.x_values[:len_t],saw_data.y_values[len_t:], saw_data.x_values[len_t:], 10, values, len_t,len_t*3, False)
#CLASSIFIER GAN
#train_data_classifier
train_data_classifier_sin = np.concatenate((curves_g_sin[:len_t*2], y_out_sin[:len_t*2]), axis=1)
train_data_classifier_saw = np.concatenate((curves_g_saw[:len_t*2], y_out_saw[:len_t*2]), axis=1)
train_data_classifier = np.concatenate((train_data_classifier_sin, train_data_classifier_saw), axis=0)
#test_data_classifier
test_data_classifier_sin = np.concatenate((curves_g_sin[len_t*2:], y_out_sin[len_t*2:]), axis=1)
test_data_classifier_saw = np.concatenate((curves_g_saw[len_t*2:], y_out_saw[len_t*2:]), axis=1)
test_data_classifier = np.concatenate((test_data_classifier_sin, test_data_classifier_saw), axis=0)
#classifier train on gan
classificatore = classifier(values)
cl = classificatore.create_clas()
num_rounds = 400
label_x = [[int(i)] for i in np.zeros(len_t*2)]
label_y = [[int(i)] for i in np.ones(len_t*2)]
labels = np.concatenate((label_x, label_y), axis=0)
label_x_test = [[int(i)] for i in np.zeros(int(len_t))]
label_y_test = [[int(i)] for i in np.ones(int(len_t))]
labels_test = np.concatenate((label_x_test, label_y_test), axis=0)
acc_class = 0
print(train_data_classifier.shape, labels.shape)
while acc_class < 0.1:
classificatore.train(train_data_classifier, labels, cl, num_rounds)
print(test_data_classifier.shape, labels_test.shape)
acc_class = classificatore.valuate(test_data_classifier, labels_test, cl)['accuracy']
print(acc_class)
num_rounds += 100
results['accuracy_classifier'] = float(acc_class)
#CLASSIFIER REAL
#train_data_classifier
sin_data = sample_data(
'sin', n_samples=len_t*3, no_drop=values, reuse=False, mul_range=[1, 2])
sin_data.create_points()
#create saw_data
saw_data = sample_data('saw', n_samples=len_t*3, mul_range=[
1, 2], no_drop=values, pred=1)
saw_data.create_points()
train_data_classifier_sin = np.concatenate((sin_data.y_values[:len_t*2], sin_data.x_values[:len_t*2]), axis=1)
train_data_classifier_saw = np.concatenate((saw_data.y_values[:len_t*2],saw_data.x_values[:len_t*2]), axis=1)
train_data_classifier = np.concatenate((train_data_classifier_sin, train_data_classifier_saw), axis=0)
#test_data_classifier
test_data_classifier_sin = np.concatenate((sin_data.y_values[len_t*2:], sin_data.x_values[len_t*2:]), axis=1)
test_data_classifier_saw = np.concatenate((saw_data.y_values[len_t*2:], saw_data.x_values[len_t*2:]), axis=1)
test_data_classifier = np.concatenate((test_data_classifier_sin, test_data_classifier_saw), axis=0)
#classifier train on gan
classificatore_real = classifier(values)
cl_real = classificatore_real.create_clas()
num_rounds = 400
label_x = [[int(i)] for i in np.zeros(len_t*2)]
label_y = [[int(i)] for i in np.ones(len_t*2)]
labels = np.concatenate((label_x, label_y), axis=0)
label_x_test = [[int(i)] for i in np.zeros(int(len_t))]
label_y_test = [[int(i)] for i in np.ones(int(len_t))]
labels_test = np.concatenate((label_x_test, label_y_test), axis=0)
acc_class = 0
print(train_data_classifier.shape, labels.shape)
while acc_class < 0.1:
classificatore_real.train(train_data_classifier, labels, cl_real, num_rounds)
print(test_data_classifier.shape, labels_test.shape)
acc_class = classificatore_real.valuate(test_data_classifier, labels_test, cl_real)['accuracy']
print(acc_class)
num_rounds += 100
results['accuracy_classifier_real'] = float(acc_class)
#create new curves sin
#create sin_data
sin_data = sample_data(
'sin', n_samples=sample, no_drop=values, reuse=False, mul_range=[1, 2])
sin_data.create_points()
#create saw_data
saw_data = sample_data('saw', n_samples=sample, mul_range=[
1, 2], no_drop=values, pred=1)
saw_data.create_points()
#train_data_sin_
test_data_sin = np.concatenate((sin_data.y_values, sin_data.x_values), axis=1)
#train_data_saw_
test_data_saw = np.concatenate((saw_data.y_values, saw_data.x_values), axis=1)
test_data = np.concatenate((test_data_sin, test_data_saw), axis=0)
#evaluate classifier on real curve
acc_class_real = classificatore.valuate(test_data, labels_test, cl)['accuracy']
acc_class_real_real = classificatore_real.valuate(test_data, labels_test, cl_real)['accuracy']
results['accuracy_real'] = float(acc_class_real)
results['accuracy_real_real'] = float(acc_class_real_real)
results['loss_sin'] = [g_loss_sin, d_loss_sin]
results['loss_saw'] = [g_loss_saw, d_loss_saw]
out_file_ac = 'results.s{}.sam{}.noise{}.json'.format(seed, sample_sizes, noise_sizes)
open(out_file_ac, 'w').write(json.dumps(results))
out_file_curves = 'curves.s{}.sam{}.noise{}.json'.format(seed, sample_sizes[0], noise_sizes[0])
open(out_file_curves, 'wb').write(pickle.dumps([[curves_g_sin,y_out_sin], [curves_g_saw,y_out_saw]]))
return results
def main():
# sample_sizes = [int(sys.argv[1]), int(sys.argv[2])]
sample_0 = 27
sample_1 = 93
values = 96
size = 32
num = int(values / size)
seed = 4
np.random.seed(seed)
results = {}
tf.reset_default_graph()
sess = tf.InteractiveSession()
#create sin_data
data = json.loads(open('ecg_data.json').read())
curve_0 = np.asarray(data['-1'])
curve_0_x = curve_0
curve_0_y = np.asarray([np.arange(values)] * sample_0)
# curve_0_x, curve_0_y =
#sample_curves(curve_0, values) #random sampling curves
#create saw_data
curve_1 = np.asarray(data['1'])
curve_1_x = curve_1
curve_1_y = np.asarray([np.arange(values)] * sample_1)
#curve_1_x, curve_1_y =
#sample_curves(curve_1, values)
slice_x_0, slice_y_0 = create_slices(curve_0_x, size)
slice_x_1, slice_y_1 = create_slices(curve_1_x, size)
print('sl', slice_x_0.shape, slice_x_1.shape)
train_x_0 = slice_x_0[:num * sample_0]
train_y_0 = slice_y_0[:num * sample_0]
train_x_1 = slice_x_1[:num * sample_1]
train_y_1 = slice_y_1[:num * sample_1]
print('k', train_x_0.shape, train_x_1.shape)
#CLASSIFIER ON GAN VALUES
#train_x_1 = curve_1_x[:sample_1]
train_data_classifier_curve_0 = np.concatenate((train_x_0, train_y_0),
axis=1)
train_data_classifier_curve_1 = np.concatenate((train_x_1, train_y_1),
axis=1)
train_data_classifier = np.concatenate(
(train_data_classifier_curve_0, train_data_classifier_curve_1), axis=0)
classificatore = classifier(size)
cl = classificatore.create_clas()
num_rounds = 900
label_x = [[int(i)]
for i in np.zeros(train_data_classifier_curve_0.shape[0])]
label_y = [[int(i)]
for i in np.ones(train_data_classifier_curve_1.shape[0])]
labels = np.concatenate((label_x, label_y), axis=0)
classificatore.train(train_data_classifier, labels, cl, num_rounds)
curve_0_y_ = np.asarray(
[np.arange(size)] *
(slice_x_0.shape[0] - train_data_classifier_curve_0.shape[0]))
curve_1_y_ = np.asarray(
[np.arange(size)] *
(slice_x_1.shape[0] - train_data_classifier_curve_1.shape[0]))
print('u', slice_x_1[sample_1 * num:sample_1 * num + 40 * num].shape,
curve_1_y_.shape)
# print(curve_1_y_.shape, slice_x_1[sample_1*num:sample_0*num+40*num].shape)
test_data_classifier_curve_0 = np.concatenate(
(slice_x_0[sample_0 * num:sample_0 * num + 40 * num], curve_0_y_),
axis=1)
test_data_classifier_curve_1 = np.concatenate(
(slice_x_1[sample_1 * num:sample_1 * num + 40 * num], curve_1_y_),
axis=1)
test_data = np.concatenate(
(test_data_classifier_curve_0, test_data_classifier_curve_1), axis=0)
label_x_test = [[int(i)] for i in np.zeros(40 * num)]
label_y_test = [[int(i)] for i in np.ones(40 * num)]
labels_test = np.concatenate((label_x_test, label_y_test), axis=0)
#evaluate classifier on real curve
acc_class_real = list(classificatore.epredict(test_data, labels_test, cl))
precision = 0
count = 0
for k in range(len(acc_class_real)):
if int(acc_class_real[k]['classes']) == 1:
count += 1
if (k + 1) % 3 == 0:
if count >= 2 and labels[k] == 1:
precision += 1
if count < 2 and labels[k] == 0:
precision += 1
count = 0
print(precision)
# acc_class_real = classificatore.valuate(test_data, labels_test, cl)['accuracy']
# acc_class_real_real = classificatore_real.valuate(test_data, labels_test, cl_real)['accuracy']
results['accuracy_real'] = precision / (len(acc_class_real) / 3)
#results['accuracy_real_real'] = float(acc_class_real_real)
# results['loss_sin'] = [g_loss_sin, d_loss_sin]
# results['loss_saw'] = [g_loss_saw, d_loss_saw]
out_file_ac = 'results.s{}.sam{}.noise{}.json'.format(
seed, sample_0, values)
open(out_file_ac, 'w').write(json.dumps(results))
# out_file_curves = 'curves.s{}.sam{}.noise{}.json'.format(seed, sample_0, values)
# open(out_file_curves, 'wb').write(pickle.dumps([[curves_0_g_out,t_0_g_out], [curves_1_g_out,t_1_g_out]]))
return results