def datasets():
my_datasets = []
anomaly_names = ["bandwidth", "download", "packet_loss"]
node_names = ["cassandra", "bono", "sprout"]
for n in node_names:
my_data = []
for idx, a in enumerate(anomaly_names):
for _ in range(3):
my_data.append(
RawData(x=np.random.rand(50, 10),
y=idx,
headers=[],
node_name=n,
anomaly_name=a))
dataset = AnomalyDataset(n,
anomaly_file_paths={},
ref_header=["blocker"])
dataset.data = my_data
my_datasets.append(dataset)
return my_datasets
def to_tensor(data):
x, y = data.x, data.y
x = torch.from_numpy(x).float().to(self.device)
y = torch.from_numpy(np.array([y])).long().to(self.device)
return RawData.create_from_ref(data, x=x, y=y)
def arr_norm(data):
x = data.x
nom = (x - self.arr_min) * (self.target_max - self.target_min)
x = self.target_min + nom / self.denom
return RawData.create_from_ref(data, x=x)
def kth_difference(data):
x = data.x
my_list = []
for i in range(self.k):
idx = i + 1
diff = x[idx:] - x[:-idx]
if self.k > 1 and idx < self.k:
diff = diff[:-(self.k - idx)]
my_list.append(diff)
x = np.stack(my_list).mean(0)
if self.append:
seq_length = min(x.shape[0], data.x.shape[0])
x = np.concatenate([x[:seq_length], data.x[:seq_length]],
axis=1)
return RawData.create_from_ref(data, x=x)
def train_list():
arr = np.random.rand(200, 10)
return [RawData(x=arr, y=1)]
def tensor_whiten(data):
x = data.x
x = np.dot(x, self.sigma_neg_sqrt)
return RawData.create_from_ref(data, x=x)
def log_scale(data):
x = data.x
x = self.log_func(x + 1) # Prevent negative infinity by adding 1
return RawData.create_from_ref(data, x=x)
def arr_stand(data):
x = data.x
x = (x - self.arr_mean) / self.arr_std
return RawData.create_from_ref(data, x=x)