def apply(self):
filename = '%s_%s' % (self.type, self.target)
cache = load_hickle_file(
os.path.join(self.settings['cachedir'], filename))
if cache is not None:
return cache
data = self.read_raw_signal()
if self.settings['dataset'] in ['CHBMIT', 'FB']:
X, y = self.preprocess(data)
elif self.settings['dataset'] in ['Kaggle2014Det']:
X, y = self.preprocess_Kaggle2014Det(data)
save_hickle_file(os.path.join(self.settings['cachedir'], filename),
[X, y])
return X, y
def apply(self, save_STFT=False, dir=''):
filename = '%s_%s' % (self.type, self.target)
cache = load_hickle_file(
os.path.join(self.settings['cachedir'] + '_ACS', filename))
if cache is not None:
return cache
data = self.read_raw_signal()
if self.settings['dataset'] == 'Kaggle2014Pred':
X, y = self.preprocess_Kaggle(data)
else:
X, y = self.preprocess(data)
save_hickle_file(
os.path.join(self.settings['cachedir'] + '_ACS', filename), [X, y])
return X, y
def apply(self, save_STFT=False, dir=''):
def save_STFT_to_files(X):
print('Start saving STFT to %s' % dir)
pre = None # oversampling for GAN training
ovl_pct = 0.1 # oversampling for GAN training
if isinstance(X, list):
index = 0
ovl_len = int(ovl_pct *
X[0].shape[-2]) # oversampling for GAN training
ovl_num = int(np.floor(1.0 / ovl_pct) -
1) # oversampling for GAN training
for x in X:
for i in range(x.shape[0]):
fn = '%s_%s_%d_%d.npy' % (self.target, self.type,
index, i)
if self.settings['dataset'] in ['FB', 'CHBMIT']:
x_ = x[i, :, :56, :112]
elif self.settings['dataset'] == 'Kaggle2014Pred':
if 'Dog' in self.target:
x_ = x[i, :, :56, :96]
elif 'Patient' in self.target:
x_ = x[i, :, :112, :96]
elif self.settings['dataset'] == 'EpilepsiaSurf':
x_ = x[i, :, :, :]
np.save(os.path.join(dir, fn), x_)
# 2352, 1, 16, 59, 100 - Kaggle Dog
# (567, 1, 15, 119, 100) - Kaggle Patient
# 3020, 1, 6, 59, 114 - Freiburg
# Generate overlapping samples for GAN
if i > 0:
for j in range(1, ovl_num + 1):
fn = '%s_ovl_%s_%d_%d_%d.npy' % (
self.target, self.type, index, i, j)
x_2 = np.concatenate((pre[:, :j * ovl_len, :],
x_[:, j * ovl_len:, :]),
axis=1)
assert x_2.shape == x_.shape
np.save(os.path.join(dir, fn), x_2)
pre = x_
index += 1
else:
ovl_len = int(ovl_pct *
X.shape[-2]) # oversampling for GAN training
ovl_num = np.floor(
1.0 / ovl_pct) - 1 # oversampling for GAN training
for i in range(X.shape[0]):
fn = '%s_%s_0_%d.npy' % (self.target, self.type, i)
if self.settings['dataset'] in ['FB', 'CHBMIT']:
x_ = X[i, :, :56, :112]
elif self.settings['dataset'] == 'Kaggle2014Pred':
if 'Dog' in self.target:
x_ = X[i, :, :56, :96]
elif 'Patient' in self.target:
x_ = X[i, :, :112, :96]
np.save(os.path.join(dir, fn), x_)
# Generate overlapping samples for GAN
if i > 0:
for j in range(1, ovl_num + 1):
fn = '%s_ovl_%s_%d_%d_%d.npy' % (
self.target, self.type, index, i, j)
x_2 = np.concatenate(
(pre[:, :j * ovl_len, :], x_[:,
j * ovl_len:, :]),
axis=-1)
assert x_2.shape == x_.shape
np.save(os.path.join(dir, fn), x_2)
pre = x_
print('Finished saving STFT to %s' % dir)
return None
filename = '%s_%s' % (self.type, self.target)
cache = load_hickle_file(
os.path.join(self.settings['cachedir'], filename))
if cache is not None:
if save_STFT:
X, _ = cache
return save_STFT_to_files(X)
else:
return cache
data = self.read_raw_signal()
if self.settings['dataset'] == 'Kaggle2014Pred':
X, y = self.preprocess_Kaggle(data)
else:
X, y = self.preprocess(data)
save_hickle_file(os.path.join(self.settings['cachedir'], filename),
[X, y])
if save_STFT:
return save_STFT_to_files(X)
else:
return X, y