Merge pull request #25 from DeepRank/class

fix classification tasks

graphprot/Metrics.py

@@ -8,7 +8,8 @@
 
 def get_boolean(values, threshold, target):
 
-    if target == 'fnat' or target == 'bin':
+    inverse = ['fnat', 'bin'] 
+    if target in inverse:
         values_bool = [1 if x > threshold else 0 for x in values]
     else:
         values_bool = [1 if x < threshold else 0 for x in values]
@@ -120,13 +121,13 @@ def hitrate(self):
 
         idx = np.argsort(self.prediction)
 
-        if self.target == 'fnat' or self.target == 'bin':
+        inverse = ['fnat', 'bin']
+        if self.target in inverse:   
             idx = idx[::-1]
 
         ground_truth_bool = get_boolean(
             self.y, self.threshold, self.target)
         ground_truth_bool = np.array(ground_truth_bool)
 
         hitrate = np.cumsum(ground_truth_bool[idx])
-
         return hitrate

graphprot/NeuralNet.py

@@ -79,7 +79,6 @@ def __init__(self, database, Net,
                 0).num_features).to(self.device)
 
         elif self.task == 'class':
-            self.classes = classes
             self.classes_idx = {i: idx for idx,
                                 i in enumerate(self.classes)}
             self.output_shape = len(self.classes)
@@ -335,19 +334,21 @@ def print_epoch_data(stage, epoch, loss, acc, time):
         print('Epoch [%04d] : %s loss %e | accuracy %s | time %1.2e sec.' % (epoch,
                                                                              stage, loss, acc_str, time))
 
-    def format_output(self, out, target):
+    def format_output(self, pred, target):
         """Format the network output depending on the task (classification/regression)."""
 
-        if self.task == 'class':
-            out = F.softmax(out, dim=1)
+
+        if self.task == 'class' :
+            pred = F.softmax(pred, dim=1)
             target = torch.tensor(
                 [self.classes_idx[int(x)] for x in target])
 
         else:
-            out = out.reshape(-1)
-
-        return out, target
+            pred = pred.reshape(-1)
+            
+        return pred, target
 
+
     def test(self, database_test, threshold=4, hdf5='test_data.hdf5'):
         """Test the model
 
@@ -380,6 +381,7 @@ def test(self, database_test, threshold=4, hdf5='test_data.hdf5'):
 
         self.test_out = _out
         self.test_y = _y
+
         _test_acc = self.get_metrics('test', threshold).accuracy
         self.test_acc = _test_acc
         self.test_loss = _test_loss
@@ -412,17 +414,23 @@ def eval(self, loader):
 
             y += d.y
             loss_val += loss_func(pred, d.y).detach().item()
-            out += pred.reshape(-1).tolist()
 
             # get the outputs for export
-            data['outputs'] += pred.reshape(-1).tolist()
+            if self.task == 'class':
+                pred = np.argmax(pred.detach(), axis=1)
+            else:
+                pred = pred.detach().reshape(-1)
+
+            out += pred
             data['targets'] += d.y.numpy().tolist()
+            data['outputs'] += pred.tolist()
 
             # get the data
             data['mol'] += d['mol']
 
         return out, y, loss_val, data
 
+
     def _epoch(self, epoch):
         """Run a single epoch
 
@@ -446,13 +454,18 @@ def _epoch(self, epoch):
             loss = self.loss(pred, d.y)
             running_loss += loss.detach().item()
             loss.backward()
-            out += pred.reshape(-1).tolist()
             self.optimizer.step()
 
             # get the outputs for export
-            data['outputs'] += pred.reshape(-1).tolist()
-            data['targets'] += d.y.numpy().tolist()
+            if self.task == 'class':
+                pred = np.argmax(pred.detach(), axis=1)
+            else:
+                pred = pred.detach().reshape(-1)
 
+            out += pred
+            data['targets'] += d.y.numpy().tolist()
+            data['outputs'] += pred.tolist()
+
             # get the data
             data['mol'] += d['mol']