jhnine
/
JCC-DeepOD
forked from JointCloud/JCC-DeepOD
0
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fix bugs in test files

This commit is contained in:
xuhongzuo 2023-11-10 12:22:23 +08:00
parent a1c3f6e893
commit dbede6f294
4 changed files with 56 additions and 50 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

69
deepod/metrics/_anomaly_detection.py
View File

@ -1,5 +1,8 @@
from sklearn import metrics
import numpy as np
from deepod.metrics.affiliation.generics import convert_vector_to_events
from deepod.metrics.vus.metrics import get_range_vus_roc
from deepod.metrics.affiliation.metrics import pr_from_events
def auc_roc(y_true, y_score):
@ -38,25 +41,42 @@ def get_best_f1(label, score):
return best_f1, best_p, best_r
# revised by Yiyuan Yang 2023/11/08
# Compored with ts_metrics, this function can return more metrics with one more intput y_test(predictions of events)
def ts_metrics_enhanced(y_true, y_score, y_test):
"""
Use Case Demo:
from deepod.models.time_series import DCdetector
clf = DCdetector()
clf.fit(X_train)
pred, scores = clf.decision_function(X_test)
from deepod.metrics import point_adjustment
from deepod.metrics import ts_metrics_enhanced
adj_eval_metrics = ts_metrics_enhanced(labels, point_adjustment(labels, scores), pred)
print('adj_eval_metrics',adj_eval_metrics)
Compared with ts_metrics, this function can return more metrics
with one more input y_test (predictions of events)
revised by @Yiyuan Yang 2023/11/08
Args:
y_true:
y_score:
y_test
Returns:
auroc:
aupr:
best_f1:
best_p:
best_r:
affiliation_precision:
affiliation_recall:
vus_r_auroc:
vus_r_aupr:
vus_roc:
vus_pr:
Example:
from deepod.models.time_series import DCdetector
clf = DCdetector()
clf.fit(X_train)
pred, scores = clf.decision_function(X_test)
from deepod.metrics import point_adjustment
from deepod.metrics import ts_metrics_enhanced
adj_eval_metrics = ts_metrics_enhanced(labels, point_adjustment(labels, scores), pred)
print('adj_eval_metrics',adj_eval_metrics)
"""
from .affiliation.generics import convert_vector_to_events
from .vus.metrics import get_range_vus_roc
from .affiliation.metrics import pr_from_events
best_f1, best_p, best_r = get_best_f1(y_true, y_score)
events_pred = convert_vector_to_events(y_test)
@ -64,7 +84,20 @@ def ts_metrics_enhanced(y_true, y_score, y_test):
Trange = (0, len(y_test))
affiliation = pr_from_events(events_pred, events_gt, Trange)
vus_results = get_range_vus_roc(y_score, y_true, 100) # default slidingWindow = 100
return auc_roc(y_true, y_score), auc_pr(y_true, y_score), best_f1, best_p, best_r, affiliation['Affiliation_Precision'], affiliation['Affiliation_Recall'], vus_results["R_AUC_ROC"], vus_results["R_AUC_PR"], vus_results["VUS_ROC"], vus_results["VUS_PR"]
auroc = auc_roc(y_true, y_score)
aupr = auc_pr(y_true, y_score)
affiliation_precision = affiliation['Affiliation_Precision']
affiliation_recall = affiliation['Affiliation_Recall']
vus_r_auroc = vus_results["R_AUC_ROC"]
vus_r_aupr = vus_results["R_AUC_PR"]
vus_roc = vus_results["VUS_ROC"]
vus_pr = vus_results["VUS_PR"]
return auroc, aupr, best_f1, best_p, best_r, \
affiliation_precision, affiliation_recall, \
vus_r_auroc, vus_r_aupr, \
vus_roc, vus_pr

2
deepod/models/time_series/couta.py
View File

@ -627,7 +627,7 @@ class _SubseqData(Dataset):
class _DSVDDUncLoss(torch.nn.Module):
def __init__(self, c, reduction='mean'):
super(DSVDDUncLoss, self).__init__()
super(_DSVDDUncLoss, self).__init__()
self.c = c
self.reduction = reduction

31
deepod/models/time_series/dcdetector.py
View File

@ -80,7 +80,6 @@ class DCdetector(BaseDeepAD):
return loss_final_pad
def training(self, dataloader):
loss_list = []
@ -161,11 +160,8 @@ class DCdetector(BaseDeepAD):
test_energy = np.array(attens_energy) # anomaly scores
return test_energy, preds # (n,d)
def predict(self, X, return_confidence=True):
seqs = get_sub_seqs(X, seq_len=self.seq_len, stride=1)
dataloader = DataLoader(seqs, batch_size=self.batch_size,
shuffle=False, drop_last=True)
@ -237,7 +233,6 @@ class DCdetector(BaseDeepAD):
test_energy = np.array(attens_energy) # anomaly scores
preds = (test_energy > thresh).astype(int)
loss_final = np.mean(test_energy, axis=1) # (n,)
loss_final_pad = np.hstack([0 * np.ones(X.shape[0] - loss_final.shape[0]), loss_final])
@ -245,33 +240,11 @@ class DCdetector(BaseDeepAD):
preds_final_pad = np.hstack([0 * np.ones(X.shape[0] - preds_final.shape[0]), preds_final])
if return_confidence:
return loss_final_pad, preds_final_pad
confidence = self._predict_confidence(loss_final_pad)
return preds_final_pad, confidence
else:
return preds_final_pad
def predict(self, X, return_confidence=False):
## self.threshold
self.threshold_ = None
# ------------------------------ #
pred_score = self.decision_function(X)
prediction = (pred_score > self.threshold_).astype('int').ravel()
if return_confidence:
confidence = self._predict_confidence(pred_score)
return prediction, confidence
return prediction
def training_forward(self, batch_x, net, criterion):
"""define forward step in training"""
return

4
deepod/test/test_dcdetector.py
View File

@ -33,8 +33,8 @@ class TestDCdetector(unittest.TestCase):
device = 'cuda' if torch.cuda.is_available() else 'cpu'
self.clf = DCdetector(seq_len=90, stride=1, epochs=2,
batch_size=32, lr=1e-4, patch_size=[5],
device=device, random_state=42)
batch_size=32, lr=1e-4, patch_size=[5],
device=device, random_state=42)
self.clf.fit(self.Xts_train)
def test_parameters(self):