def get(data):
# inits
train_index, val_index, test_index = (
[None] * folds,
[None] * folds,
[None] * folds,
)
groups = [None] * folds
group = stringlist2ind([k for k in data[group_key]])
for k in np.unique(group):
tmp_split = np.array_split(np.where(group == k)[0], folds)
for f in range(folds):
if k == 0:
groups[f] = np.empty(0)
groups[f] = np.append(groups[f], tmp_split[f]).astype(int)
for f in range(folds):
sel = np.setdiff1d(np.arange(folds), f)
train_index[f] = listnp_combine([groups[k] for k in sel])
val_inds = np.random.choice(
range(len(train_index[f])),
int(np.ceil(len(train_index[f]) * val_frac)),
replace=False,
)
val_index[f] = train_index[f][val_inds]
train_index[f] = train_index[f][
list(set(range(len(train_index[f]))) - set(val_inds))
]
return {"train": train_index, "val": val_index, "test": groups}
def _get_binary_anomaly(self, paths):
subdbs = np.unique(self['data']['subdb'])
labels = [None] * len(subdbs)
for k, subdb in enumerate(subdbs):
subdb_id = np.where([s == subdb for s in self['data']['subdb']])[0]
reorder = np.array([int(os.path.splitext(filename)[0]) \
for filename in \
[self['data']['filename'][k] for k in subdb_id]])
labels[k] = np.load(
os.path.join(paths['meta'], subdb + '_labels.npy'))[reorder]
return listnp_combine(labels)
def process(self, data, **kwargs):
# aggregate data
tmp = [None] * len(self.methods)
for k in range(len(self.methods)):
if self.methods[k] == 'mean':
tmp[k] = np.mean(data, axis=self.axis)
elif self.methods[k] == 'std':
tmp[k] = np.std(data, axis=self.axis)
elif self.methods[k] == 'kurtosis':
tmp[k] = scipy.stats.kurtosis(data, axis=self.axis)
else:
print('Aggregation method not supported')
sys.exit()
output = (listnp_combine(tmp, method=self.combine, axis=self.combine_axis) if self.combine is not None else tmp)
return output, {'time_step': 0} if self.axis == 0 else dict()
def process(self, data: np.ndarray, **kwargs) -> (np.ndarray, Dict):
# aggregate data
tmp = [None] * len(self.methods)
for k in range(len(self.methods)):
if self.methods[k] == "mean":
tmp[k] = np.mean(data, axis=self.axis)
elif self.methods[k] == "std":
tmp[k] = np.std(data, axis=self.axis)
elif self.methods[k] == "kurtosis":
tmp[k] = scipy.stats.kurtosis(data, axis=self.axis)
else:
print("Aggregation method not supported")
sys.exit()
output = (listnp_combine(
tmp, method=self.combine, axis=self.combine_axis)
if self.combine is not None else tmp)
return output, {"time_step": 0} if self.axis == 0 else dict()