undo

web_tool/ServerModelsTorchSmoothingCycle.py

@@ -65,6 +65,7 @@ class TorchSmoothingCycleFineTune(BackendModel):
         
         self.corr_features = [[] for _ in range(num_models) ]
         self.corr_labels = [[] for _ in range(num_models) ]
+        self.num_corrections_since_retrain = [ [ 0 for _ in range(num_models) ] ]
 
     def run(self, naip_data, naip_fn, extent):
         print(naip_data.shape)
@@ -191,11 +192,15 @@ class TorchSmoothingCycleFineTune(BackendModel):
         return success, message
     
     def undo(self):
-        pass
-        #if len(self.corr_features)>0:
-        #    self.corr_features = self.corr_features[:-1]
-        #    self.corr_labels = self.corr_labels[:-1]
-        #print('undoing; now there are %d samples' % len(self.corr_features))
+        num_undone = sum(self.num_corrections_since_retrain[-1])
+        message = 'Removed {} labels'.format(' '.join(map(str,self.num_corrections_since_retrain[-1])))
+        for i in range(self.num_models):
+            self.corr_features[i] = self.corr_features[i][:-self.num_corrections_since_retrain[-1][i]]
+            self.corr_labels[i] = self.corr_labels[i][:-self.num_corrections_since_retrain[-1][i]]
+        self.num_corrections_since_retrain = self.num_corrections_since_retrain[:-1]
+        if len(self.num_corrections_since_retrain) == 0:
+            self.num_corrections_since_retrain = [ [ 0 for _ in range(self.num_models)] ]
+        return True, message, num_undone
 
     def add_sample(self, tdst_row, bdst_row, tdst_col, bdst_col, class_idx, model_idx):
         print("adding sample: class %d (incremented to %d) at (%d, %d), model %d" % (class_idx, class_idx+1 , tdst_row, tdst_col, model_idx))
@@ -204,6 +209,7 @@ class TorchSmoothingCycleFineTune(BackendModel):
             for j in range(tdst_col,bdst_col+1):
                 self.corr_labels[model_idx].append(class_idx+1)
                 self.corr_features[model_idx].append(self.features[0,:,i,j])
+                self.num_corrections_since_retrain[-1][model_idx] += 1
 
     def init_model(self):
         checkpoint = T.load(self.model_fn, map_location=self.device)
@@ -217,6 +223,7 @@ class TorchSmoothingCycleFineTune(BackendModel):
         
     def reset(self):
         self.init_model()
+        for i in range(self.num_models): self.num_corrections_since_retrain[i] = 0
 
     def run_core_model_on_tile(self, naip_tile):