tsadmetrics 0.1.9__py3-none-any.whl → 0.1.10__py3-none-any.whl

tests/test_non_binary.py

@@ -202,7 +202,7 @@ class TestAUCPRPA(unittest.TestCase):
         self.assertAlmostEqual(score, expected_score, places=4)
 
         score = round(auc_pr_pa(self.y_true1, self.y_pred3),2)
-        expected_score = 0.5
+        expected_score = 0.75
         self.assertAlmostEqual(score, expected_score, places=4)
 
         if len(self.y_true2) == len(self.y_pred4):
@@ -252,7 +252,7 @@ class TestAUCPRSW(unittest.TestCase):
         self.assertAlmostEqual(score, expected_score, places=4)
 
         score = round(auc_pr_sw(self.y_true1, self.y_pred3),2)
-        expected_score = 0.5
+        expected_score = 1
         self.assertAlmostEqual(score, expected_score, places=4)
 
         

tsadmetrics/non_binary_metrics.py

@@ -26,19 +26,69 @@ def auc_pr_pw(y_true : np.array ,y_anomaly_scores:  np.array):
 
 
 def auc_pr_pa(y_true: np.array, y_anomaly_scores: np.array):
-    thresholds = np.unique(y_anomaly_scores)[::-1]  # Descending order
     precisions = [1]
     recalls = [0]
-    for t in thresholds[:-1]:
+    tps,fps,fns = [],[],[]
 
-        y_pred = (y_anomaly_scores >= t).astype(int)
-
-
-        precisions.append(point_adjusted_precision(y_true, y_pred))
-        recalls.append(point_adjusted_recall(y_true, y_pred))
+    p_adj = PointAdjust(len(y_true),y_true,(np.array(y_anomaly_scores) >= 0.5).astype(int))
+    segments= p_adj.get_gt_anomalies_segmentwise()
+    idx = np.argsort(y_anomaly_scores)[::-1].astype(int)
+    y_true_sorted = np.array(y_true)[idx]
+    y_anomaly_scores_sorted = np.array(y_anomaly_scores)[idx]
+    
+    segment_mins = []
+    for start,end in segments:
+        anoms_scores = y_anomaly_scores[start:end+1]
+        segment_mins.append([np.max(anoms_scores),end-start+1])
+
+    for i_t in range(len(y_anomaly_scores_sorted)):
+        fp,tp,fn = 0,0,0
+        if i_t > 0 and y_anomaly_scores_sorted[i_t] == y_anomaly_scores_sorted[i_t-1] :
+            tp = tps[-1]
+            fp = fps[-1]
+            fn = fns[-1]
+        else:
+            if y_true_sorted[i_t] == 0:
+                #FP
+                if len(fps)==0:
+                    aux_y_pred = (y_anomaly_scores >= y_anomaly_scores_sorted[i_t]).astype(int)
+                    for i in range(len(aux_y_pred)):
+                        if aux_y_pred[i] == 1 and y_true[i] == 0:
+                            fp+=1
+
+
+                else:
+                    fp=fps[i_t-1]+1
+            else:
+                if len(fps)==0:
+                    aux_y_pred = (y_anomaly_scores >= y_anomaly_scores_sorted[i_t]).astype(int)
+                    for i in range(len(aux_y_pred)):
+                        if aux_y_pred[i] == 1 and y_true[i] == 0:
+                            fp+=1
+                else:
+                    fp=fps[i_t-1]
+            for score, length in segment_mins:
+                if score >= y_anomaly_scores_sorted[i_t]:
+                    #TP
+                    tp+= length
+                else:
+                    #FN
+                    fn+= length
+        tps.append(tp)
+        fns.append(fn)
+        fps.append(fp)
+    for tp,fp,fn in zip(tps,fps,fns):
+        if tp>0:
+            precisions.append(tp/(tp+fp))
+            recalls.append(tp/(tp+fn))
+        else:
+            precisions.append(0)
+            recalls.append(0)    
+      
     
     recalls.append(1)
     precisions.append(0)
+
     auc_value = auc(recalls, precisions)
     return auc_value
 
@@ -46,17 +96,91 @@ def auc_pr_pa(y_true: np.array, y_anomaly_scores: np.array):
 
 
 def auc_pr_sw(y_true: np.array, y_anomaly_scores: np.array):
-    thresholds = np.unique(y_anomaly_scores)[::-1]  # Descending order
     precisions = [1]
     recalls = [0]
+    tps,fps,fns = [],[],[]
+
+
+    segments = []
+    i=0
+    while i < len(y_true):
+        if y_true[i] == 1:
+            start = i
+            end = i
+            while i < len(y_true) and y_true[i] == 1:
+                end = i
+                i += 1
+            segments.append([start,end])
+        i+=1
+    idx = np.argsort(y_anomaly_scores)[::-1].astype(int)
+    y_anomaly_scores_sorted = np.array(y_anomaly_scores)[idx]
     
-    for t in thresholds[:-1]:
-        y_pred = (y_anomaly_scores >= t).astype(int)
-        precisions.append(segment_wise_precision(y_true, y_pred))
-        recalls.append(segment_wise_recall(y_true, y_pred))
-    recalls.append(1)
-    precisions.append(0)
+    segment_mins = []
+    for start,end in segments:
+        anoms_scores = y_anomaly_scores[start:end+1]
+        segment_mins.append([np.max(anoms_scores),[start,end]])
+
+    for i_t in range(len(y_anomaly_scores_sorted)):
+        fp,tp,fn = 0,0,0
+        
+            
+        aux_y_pred = (y_anomaly_scores >= y_anomaly_scores_sorted[i_t]).astype(int) 
+        for score,seg in segment_mins:
+            start,end = seg
+            if score >= y_anomaly_scores_sorted[i_t]:
+                #TP
+                tp+= 1
+                if aux_y_pred[start]== 1:
+                    # Extender hacia la izquierda
+                    i = start - 1
+                    while i >= 0 and aux_y_pred[i] == 1:
+                        aux_y_pred[i] = 0
+                        i -= 1
+                
+                if aux_y_pred[end] == 1:
+                    # Extender hacia la derecha
+                    i = end + 1
+                    while i < len(aux_y_pred) and aux_y_pred[i] == 1:
+                        aux_y_pred[i] = 0
+                        i += 1
+                aux_y_pred[start:end+1] = 0
+
+            else:
+                #FN
+                fn+= 1
+        
+        if np.sum(aux_y_pred)>0:
+            fpsegments = []
+            i=0
+            while i < len(aux_y_pred):
+                if aux_y_pred[i] == 1:
+                    start = i
+                    end = i
+                    while i < len(aux_y_pred) and aux_y_pred[i] == 1:
+                        end = i
+                        i += 1
+                    fpsegments.append([start,end])
+                i+=1
+            fp = len(fpsegments)
+        else:
+            fp = 0
+            
+            
+        tps.append(tp)
+        fns.append(fn)
+        fps.append(fp)
+    for tp,fp,fn in zip(tps,fps,fns):
+        if tp>0:
+            precisions.append(tp/(tp+fp))
+            recalls.append(tp/(tp+fn))
+        else:
+            precisions.append(0)
+            recalls.append(0)    
+      
+    
+
     auc_value = auc(recalls, precisions)
+
     return auc_value
 
 

{tsadmetrics-0.1.9.dist-info → tsadmetrics-0.1.10.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: tsadmetrics
-Version: 0.1.9
+Version: 0.1.10
 Summary: =?unknown-8bit?q?Librer=C3=ADa_para_evaluaci=C3=B3n_de_detecci=C3=B3n_de_anomal=C3=ADas?= en series temporales
 Home-page: https://github.com/pathsko/TSADmetrics
 Author: Pedro Rafael Velasco Priego

{tsadmetrics-0.1.9.dist-info → tsadmetrics-0.1.10.dist-info}/RECORD

@@ -13,11 +13,11 @@ entorno/bin/rst2xml.py,sha256=uoIfpn3prnir2tzqdycsAjOg-OWw663XOK47IeHCZdY,651
 entorno/bin/rstpep2html.py,sha256=sthYQHEgYfj4JqwG45URwVbRAs-HYuwKget7SUwp9fc,719
 tests/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 tests/test_binary.py,sha256=3uLdXzdbQcqLRfXeKDmt2g2XQTU5lZOFQoiy-r9Olqo,29801
-tests/test_non_binary.py,sha256=PRvsgGbwbRXVNkYBk2-oJw8ekLW4DWJkEZhAWO0L9sk,11309
+tests/test_non_binary.py,sha256=syANlwm0DKsL6geGeq6nQI6ZVe6T_YXWTyk2-Hmck4s,11308
 tsadmetrics/__init__.py,sha256=MTWOa43fgOdkMNo5NglCReRnB8hoF0ob2PIvDziCNHw,1575
 tsadmetrics/binary_metrics.py,sha256=nwfPdfHAc_4tJMNlyIwMwFQRLvCU-ik9lQLqlaWLqTs,37741
 tsadmetrics/metric_utils.py,sha256=Y_lOE01_uyC22wnw3_G-kKUEJdqevDIWMWvSDE8Cjms,10477
-tsadmetrics/non_binary_metrics.py,sha256=JIOvkigSjHBZLKbGJj7ESe0lPM7P1JPoIUnbiMZuuLg,2896
+tsadmetrics/non_binary_metrics.py,sha256=hmARpwaYNl_u36uOHcTZqO3nd0LkHpJjPBtbqT6yP_g,6739
 tsadmetrics/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 tsadmetrics/utils.py,sha256=fV5sJE094C_GjBbqrI34Wpy-4hcZtXc9y207ffQB7Mc,2360
 tsadmetrics/_tsadeval/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -52,7 +52,7 @@ tsadmetrics/_tsadeval/prts/time_series_metrics/fscore.py,sha256=pJz4iuPyVGNvwsaR
 tsadmetrics/_tsadeval/prts/time_series_metrics/precision.py,sha256=jLkcMg7UNl25SHtZUBGkP-RV8HsvaZCtjakryl7PFWU,3204
 tsadmetrics/_tsadeval/prts/time_series_metrics/precision_recall.py,sha256=OhUJSm_I7VZ_gX_SSg8AYUq3_NW9rMIy7lAVsnOFw4Q,417
 tsadmetrics/_tsadeval/prts/time_series_metrics/recall.py,sha256=LL-0pPer3ymovVRlktaHo5XDzpgiDhWOVfdPOzKR6og,3152
-tsadmetrics-0.1.9.dist-info/METADATA,sha256=tFW_YWxVHIkXA2qGqORTi2HGHfZ1akAE5aAKQrh7yhI,830
-tsadmetrics-0.1.9.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
-tsadmetrics-0.1.9.dist-info/top_level.txt,sha256=WHaYe-ubr_88yhxe-SaZC8HuAMvlSjXCo-wIdkTeKtA,26
-tsadmetrics-0.1.9.dist-info/RECORD,,
+tsadmetrics-0.1.10.dist-info/METADATA,sha256=qYloOTiFkW1RZqlJMvIAm6rjYg_atf4i11aF6lrCQXU,831
+tsadmetrics-0.1.10.dist-info/WHEEL,sha256=iAkIy5fosb7FzIOwONchHf19Qu7_1wCWyFNR5gu9nU0,91
+tsadmetrics-0.1.10.dist-info/top_level.txt,sha256=WHaYe-ubr_88yhxe-SaZC8HuAMvlSjXCo-wIdkTeKtA,26
+tsadmetrics-0.1.10.dist-info/RECORD,,