def view_logfile(filename, show=True):
"""Plots the data from a logfile"""
fp = open(filename, 'r')
data = json.load(fp)
original_data = data['original_data']
processed_data = data['processed_data']
results = data['results']
plot_data(filename, original_data, processed_data, results, show)
fp.close()
def view_logfile(filename, show=True):
"""Plots the data from a logfile"""
fp = open(filename, "r")
data = json.load(fp)
original_data = data["original_data"]
processed_data = data["processed_data"]
results = data["results"]
plot_data(filename, original_data, processed_data, results, show)
fp.close()
def main(filename):
fp = open(filename, 'r')
data = json.load(fp)
fp.close()
detector_data = data[0]['datapoints']
timeseries = [t[0] for t in detector_data]
SpikeDetector.convert_null_values(timeseries)
timestamps = [t[1] for t in detector_data]
orig_series = timeseries[:]
orig_stamps = timestamps[:]
res = DETECTOR.detect_anomalies(timeseries, timestamps)
plot_data(filename, (orig_stamps, orig_series), (timestamps, timeseries), res)
def test(self, X_test=None, y_test=None, split=True):
"""Online testing instance one by one
Parameters
----------
X_test:
y_test:
split: True
split each instance into subflows
Returns
-------
"""
# set model in the "eval: mode.
self.model.eval()
test_loader = zip(X_test, y_test)
# Calculate Accuracy
correct = 0
total = 0
y_score = []
outputs = []
# Iterate through test dataset one by one
for i, (X, y) in enumerate(test_loader):
if split:
new_Xs, new_ys = split_instance(X, y)
res = []
# testing subflow with adding more packets
for i, (_x, _y) in enumerate(zip(new_Xs, new_ys)):
# print(i, len(X))
_x = Tensor(_x)
_y = Tensor([_y])
if torch.cuda.is_available():
_x = Variable(
_x.view(1, _x.shape[0], self.in_dim).cuda())
else:
_x = Variable(_x.view(1, _x.shape[0], self.in_dim))
# Forward pass only to get logits/output
if torch.cuda.is_available():
v = torch.sum(
(self.model(_x).cpu() - self.centers.cpu())**2,
dim=1).data.numpy()
else:
v = torch.sum((self.model(_x) - self.centers)**2,
dim=1).data.numpy()
res.append(v.item(
)) # [subflow1_res, subflow2_res,...] (for each flow)
outputs.append(res)
results = []
R = np.percentile(self.normal_dists, q=90)
succeeded_flows = []
failed_flows = []
print("len of outputs:", len(outputs))
with open(r'D:\研一\异常流量检测\result.txt', 'a') as f:
for i in outputs:
f.write(str(i) + '\n')
#print("outputs:",outputs)
for vs in outputs:
find_flg = False
_scores = []
k = 1
v_b = 0
for j, v in enumerate(
vs
): # res := [subflow1_res, subflow2_res,...] (for each flow)
_score, _confidence = self.score_function(v)
_scores.append([_score, _confidence])
###########这里粗暴break有问题
# if v > R:
# find_flg = True
# break
v_b = v_b + v * k
#print(v)
#print(torch.sigmoid(torch.Tensor([v_b])))
print(torch.Tensor([v_b]))
if torch.sigmoid(torch.Tensor([v_b])) <= 0.1:
find_flg = True
break
k += 1
#print(k)
res = {
'needed_ptks': j + 1,
'R': R,
'prob': v,
'find_flg': find_flg,
'time': 0,
'_scores': _score
}
results.append(res)
if res['find_flg']:
succeeded_flows.append(res['needed_ptks'])
else:
failed_flows.append(res['needed_ptks'])
if len(succeeded_flows) > 0:
print(
f'average_needed_pkts: {np.mean(succeeded_flows)} +/- {np.std(succeeded_flows)} '
f'when R: {R}, succeeded_flows: {len(succeeded_flows)}, all_flows: {len(y_test)}'
)
if len(failed_flows) > 0:
print(
f'failed_flows: {np.mean(failed_flows)} +/- {np.std(failed_flows)} when R: {R}, '
f'failed_flows: {len(failed_flows)}, all_flows: {len(y_test)}')
# plot the figure of num of packets and aucs
min_flow_len = min([len(vs) for vs in outputs])
print(f'***min_flow_len: {min_flow_len}')
aucs = []
accs = []
part_results = np.asarray([v[:min_flow_len] for v in outputs])
for i in range(min_flow_len):
fpr, tpr, _ = roc_curve(y_test, part_results[:, i], pos_label=1)
auc = metrics.auc(fpr, tpr)
aucs.append(auc)
y_pred = [1 if v < self.R else 0 for v in part_results[:, i]]
acc = metrics.accuracy_score(y_test, y_pred)
print(acc)
accs.append(acc)
print(accs)
plot_data(range(1, min_flow_len + 1),
aucs,
xlabel='num of packets in each flow',
ylabel='auc',
title='')
plot_data(range(1, min_flow_len + 1),
accs,
xlabel='num of packets in each flow',
ylabel='accuracy',
title='')
print(f'aucs: {aucs}')
print(f'accs: {accs}')
