def evaluate():
name = "final"
locations = ['bag', 'hand', 'hips', 'torso']
datasets = ['train', 'validation']
for dset in datasets:
for loc in locations:
params = get_params(name=name,
FX_sel='basic',
location=loc,
dset_L='validation',
dset_U='test',
dset_V=dset,
ratio_L=1.0,
ratio_U=1.0,
ratio_V=1.0,
prediction=False,
evaluate=True,
pretrain='pretrain',
runs=10,
epochs=50,
save_step=2,
oversampling=True,
G_label_sample=True,
G_label_factor=1,
C_basic_train=True,
R_active=False,
G_no=1,
D_no=1,
C_no=1,
log_name='log')
train(params=params)
def basic():
name = "final"
params = get_params(name=name,
FX_sel='basic',
location='hips',
dset_L='validation',
dset_U='test',
dset_V='validation',
ratio_L=1.0,
ratio_U=1.0,
ratio_V=1.0,
prediction=True,
pretrain='pretrain',
runs=20,
epochs=50,
save_step=2,
oversampling=True,
G_label_sample=True,
G_label_factor=1,
C_basic_train=True,
R_active=False,
G_no=1,
D_no=1,
C_no=1,
log_name='log')
train(params=params)
def test():
name = "test"
network.clear(name)
params = get_params(name=name,
FX_sel='basic',
location='hips',
dset_L='train',
dset_U='validation',
dset_V='validation',
ratio_L=1,
ratio_U=1,
ratio_V=1,
prediction=True,
pretrain='final',
runs=1,
epochs=10,
save_step=2,
oversampling=True,
G_label_sample=True,
G_label_factor=1,
C_basic_train=True,
R_active=True,
G_no=1,
D_no=1,
C_no=1,
log_name='log')
train(params=params)
GAN.get_prediction_accuracy(params)
def load_FX(FX_sel):
from GAN import get_params
params = get_params(FX_sel=FX_sel)
params['location'] = 'all'
data, label = load_data(params, 'validation')
print('Loaded!')
print(" Data:", data.shape)
print(" NaN:", np.count_nonzero(np.isnan(data)))
print("Label:", label.shape)
print("----------------------")
params['location'] = 'bag'
data, label = load_data(params, 'train')
print('Loaded!')
print(" Data:", data.shape)
print(" NaN:", np.count_nonzero(np.isnan(data)))
print("Label:", label.shape)
print("----------------------")
params['location'] = 'hand'
data, label = load_data(params, 'train')
print('Loaded!')
print(" Data:", data.shape)
print(" NaN:", np.count_nonzero(np.isnan(data)))
print("Label:", label.shape)
print("----------------------")
params['location'] = 'hips'
data, label = load_data(params, 'train')
print('Loaded!')
print(" Data:", data.shape)
print(" NaN:", np.count_nonzero(np.isnan(data)))
print("Label:", label.shape)
print("----------------------")
params['location'] = 'torso'
data, label = load_data(params, 'train')
print('Loaded!')
print(" Data:", data.shape)
print(" NaN:", np.count_nonzero(np.isnan(data)))
print("Label:", label.shape)
print("----------------------")
params['location'] = 'test'
data, label = load_data(params, 'test')
print('Loaded!')
print(" Data:", data.shape)
print(" NaN:", np.count_nonzero(np.isnan(data)))
print("----------------------")
def cross_train(def_params,
param_1,
param_1_lst,
param_2,
param_2_lst,
last_bmark=0):
clear('results')
results = np.zeros((len(param_1_lst), len(param_2_lst), 3))
count = 1
for x, p1 in enumerate(param_1_lst):
for y, p2 in enumerate(param_2_lst):
c_name = def_params['name'] + '_' + str(count).zfill(3)
params = network.load_Parameter(c_name)
if params is None:
params = def_params.copy()
params['name'] = c_name
params[param_1] = p1
params[param_2] = p2
else:
params = get_params(**params)
# Continue at the last benchmark
if count > last_bmark:
log('Benchmark %s: %s = %s | %s = %s' %
(c_name, param_1, str(p1), param_2, str(p2)),
name='results')
train(params)
mat_accuracy_G, mat_accuracy_D, mat_accuracy_C = network.load_Acc(
params)
acc_G = np.mean(mat_accuracy_G, axis=0)
acc_D = np.mean(mat_accuracy_D, axis=0)
acc_C = np.mean(mat_accuracy_C, axis=0)
results[x, y, 1] = acc_G[-1]
results[x, y, 2] = acc_D[-1]
results[x, y, 0] = acc_C[-1]
count += 1
excel.save(def_params, param_1, param_1_lst, param_2, param_2_lst, results)
def pretrain():
name = 'pretrain'
params = get_params(name=name,
FX_sel='basic',
location='hips',
dset_L='train',
dset_V='validation',
ratio_L=1.00,
ratio_V=1.00,
runs=20,
epochs=25,
save_step=5,
oversampling=True,
C_no=1,
log_name='log')
train_C(params=params)
def cross_params(net, lst_1, lst_2, last_bmark=0):
name = 'cross_' + net
params = get_params(name=name,
FX_sel='basic',
location='hips',
dset_L='validation',
dset_U='validation',
dset_V=None,
ratio_L=0.5,
ratio_U=0.5,
runs=5,
epochs=150,
save_step=15,
oversampling=True,
G_label_sample=True,
G_label_factor=1,
C_basic_train=True,
R_active=False,
G_no=1,
D_no=1,
C_no=1,
log_name='log')
network.save_Parameter(params)
param_1 = net + 'LR'
param_1_lst = lst_1
param_2 = net + 'B1'
param_2_lst = lst_2
log('Crosstrain param %s against param %s.' % (param_1, param_2),
name=params['log_name'])
log('List 1: [%s]' % (' '.join(str(v) for v in param_1_lst)),
name=params['log_name'])
log('List 2: [%s]' % (' '.join(str(v) for v in param_2_lst)),
name=params['log_name'])
log('Total runs: %d' % (len(param_1_lst) * len(param_2_lst)),
name=params['log_name'])
cross_train(params, param_1, param_1_lst, param_2, param_2_lst, last_bmark)
def test_cross():
name = 'test_cross'
params = get_params(name=name,
FX_sel='basic',
location='hips',
dset_L='validation',
dset_U='validation',
runs=3,
epochs=6,
save_step=3,
oversampling=False,
G_label_sample=True,
G_label_factor=1,
C_basic_train=True,
R_active=False,
G_no=1,
D_no=1,
C_no=1,
ratio_L=0.05,
ratio_U=0.33,
log_name='log')
network.save_Parameter(params)
param_1 = 'GLR'
param_1_lst = [0.0005, 0.001, 0.0015]
param_2 = 'GB1'
param_2_lst = [0.3, 0.7, 0.9, 0.99]
log('Crosstrain param %s against param %s.' % (param_1, param_2),
name=params['log_name'])
log('List 1: [%s]' % (' '.join(str(v) for v in param_1_lst)),
name=params['log_name'])
log('List 2: [%s]' % (' '.join(str(v) for v in param_2_lst)),
name=params['log_name'])
log('Total runs: %d' % (len(param_1_lst) * len(param_2_lst)),
name=params['log_name'])
cross_train(params, param_1, param_1_lst, param_2, param_2_lst, 0)
def rerun():
train(params=get_params(name="final"))
train(params=get_params(name="Val_Val"))
train(params=get_params(name="Val_Train"))
