add RK3 traning script

dcase2020_task2/data_sets/abnormal_datasets.py

@@ -23,7 +23,8 @@ class ComplementMCMDataSet(BaseDataSet):
             fmin=0,
             normalize_raw=False,
             normalize=None,
-            hop_all=False
+            hop_all=False,
+            same_type=False
     ):
         self.data_root = data_root
         self.context = context
@@ -66,22 +67,22 @@ class ComplementMCMDataSet(BaseDataSet):
             validation_set.data = (validation_set.data - mean) / std
 
         training_sets = []
-        validation_sets = []
+        # validation_sets = []
 
         for type_ in TRAINING_ID_MAP:
             for id_ in TRAINING_ID_MAP[type_]:
-                if type_ != machine_type:
+                if type_ != machine_type or (id_ != machine_id and same_type):
                     t = MachineDataSet(type_, id_, mode='training', **kwargs)
                     t.data = (t.data - mean) / std
 
-                    v = MachineDataSet(type_, id_, mode='validation', **kwargs)
-                    v.data = (v.data - mean) / std
+                    #v = MachineDataSet(type_, id_, mode='validation', **kwargs)
+                    #v.data = (v.data - mean) / std
 
                     training_sets.append(t)
-                    validation_sets.append(v)
+                    # validation_sets.append(v)
 
         self.training_set = torch.utils.data.ConcatDataset(training_sets)
-        self.validation_set = torch.utils.data.ConcatDataset(validation_sets)
+        # self.validation_set = torch.utils.data.ConcatDataset(validation_sets)
         self.mean = mean
         self.std = std
 
@@ -93,7 +94,7 @@ class ComplementMCMDataSet(BaseDataSet):
         return self.training_set
 
     def validation_data_set(self):
-        return self.validation_set
+        raise NotImplementedError
 
     def mean_std(self):
         return self.mean, self.std

dcase2020_task2/experiments/classification_experiment.py

@@ -155,6 +155,9 @@ def configuration():
     num_hidden = 3
     dropout_probability = 0.0
 
+    # complement set
+    same_type = True
+
     debug = False
     if debug:
         num_workers = 0
@@ -170,7 +173,7 @@ def configuration():
     normalize_raw = True
 
     # TODO: change default descriptor
-    descriptor = "ClassificationExperiment_Model:[{}_{}_{}_{}]_Training:[{}_{}_{}_{}]_Features:[{}_{}_{}_{}_{}_{}_{}]_{}".format(
+    descriptor = "ClassificationExperiment_Model:[{}_{}_{}_{}]_Training:[{}_{}_{}_{}]_Features:[{}_{}_{}_{}_{}_{}_{}]_Complement:[{}]{}".format(
         model_class,
         hidden_size,
         num_hidden,
@@ -186,6 +189,7 @@ def configuration():
         hop_size,
         power,
         fmin,
+        same_type,
         seed
     )
 
@@ -226,7 +230,8 @@ def configuration():
             'normalize_raw': normalize_raw,
             'power': power,
             'fmin': fmin,
-            'hop_all': False
+            'hop_all': False,
+            'same_type': same_type
         }
     }
 

dcase2020_task2/losses/auc_loss.py

@@ -9,13 +9,12 @@ class AUC(BaseLoss):
         super().__init__()
         self.weight = weight
 
-    def forward(self, batch_normal, batch_abnormal):
+    def forward(self, batch_normal):
 
-        assert batch_normal.get('scores'), "cannot compute loss without scores"
-        assert batch_abnormal.get('scores'), "cannot compute loss without scores"
+        assert batch_normal.get('scores') is not None, "cannot compute loss without scores"
 
-        normal_scores = batch_normal['scores']
-        abnormal_scores = batch_abnormal['scores']
+        normal_scores = batch_normal['scores'][batch_normal['abnormal'] == 0]
+        abnormal_scores = batch_normal['scores'][batch_normal['abnormal'] == 1]
 
         tprs = torch.sigmoid(abnormal_scores[:, None] - normal_scores[None, :]).mean(dim=0)
         batch_normal['tpr'] = tprs.mean()
@@ -28,6 +27,6 @@ class AUC(BaseLoss):
         batch_normal['normal_scores_mean'] = normal_scores.mean()
         batch_normal['normal_scores_std'] = normal_scores.std()
         batch_normal['abnormal_scores_mean'] = abnormal_scores.mean()
-        batch_normal['abnormal_scores_std'] = normal_scores.std()
+        batch_normal['abnormal_scores_std'] = abnormal_scores.std()
 
-        return batch_normal[f'{self.prefix}_loss']
+        return batch_normal

dcase2020_task2/losses/np_loss.py

@@ -7,15 +7,14 @@ class NP(BaseLoss):
     def __init__(self, weight=1.0, rho=0.2, **kwargs):
         super().__init__()
         self.weight = weight
-        self.rho=0.2
+        self.rho = 0.2
 
-    def forward(self, batch_normal, batch_abnormal, *args, **kwargs):
+    def forward(self, batch_normal, *args, **kwargs):
 
-        assert batch_normal.get('scores'), "cannot compute loss without scores"
-        assert batch_abnormal.get('scores'), "cannot compute loss without scores"
+        assert batch_normal.get('scores') is not None, "cannot compute loss without scores"
 
-        normal_scores = batch_normal['scores']
-        abnormal_scores = batch_abnormal['scores']
+        normal_scores = batch_normal['scores'][batch_normal['abnormal'] == 0]
+        abnormal_scores = batch_normal['scores'][batch_normal['abnormal'] == 1]
 
         with torch.no_grad():
             phi = torch.kthvalue(normal_scores, int((1 - self.rho) * normal_scores.shape[0]))[0]
@@ -30,6 +29,6 @@ class NP(BaseLoss):
         batch_normal['normal_scores_mean'] = normal_scores.mean()
         batch_normal['normal_scores_std'] = normal_scores.std()
         batch_normal['abnormal_scores_mean'] = abnormal_scores.mean()
-        batch_normal['abnormal_scores_std'] = normal_scores.std()
+        batch_normal['abnormal_scores_std'] = abnormal_scores.std()
 
         return batch_normal