bug fix

Author: Chengcheng-Guo

deepcore/methods/craig.py

@@ -53,7 +53,7 @@ def calc_gradient(self, index=None):
         for i, (input, targets) in enumerate(batch_loader):
             self.model_optimizer.zero_grad()
             outputs = self.model(input.to(self.args.device))
-            loss = self.criterion(torch.nn.functional.softmax(outputs.requires_grad_(True), dim=1),
+            loss = self.criterion(outputs.requires_grad_(True),
                                   targets.to(self.args.device)).sum()
             batch_num = targets.shape[0]
             with torch.no_grad():

deepcore/methods/glister.py

@@ -46,7 +46,7 @@ def calc_gradient(self, index=None, val=False, record_val_detail=False):
         for i, (input, targets) in enumerate(batch_loader):
             self.model_optimizer.zero_grad()
             outputs = self.model(input.to(self.args.device))
-            loss = self.criterion(torch.nn.functional.softmax(outputs.requires_grad_(True), dim=1), targets.to(self.args.device)).sum()
+            loss = self.criterion(outputs.requires_grad_(True), targets.to(self.args.device)).sum()
             batch_num = targets.shape[0]
             with torch.no_grad():
                 bias_parameters_grads = torch.autograd.grad(loss, outputs)[0]
@@ -93,7 +93,7 @@ def update_val_gradients(self, new_selection, selected_for_train):
             batch_indx = np.arange(sample_num)[i * self.args.selection_batch:min((i + 1) * self.args.selection_batch,
                                                                                  sample_num)]
             new_out_puts_batch = new_outputs[batch_indx].clone().detach().requires_grad_(True)
-            loss = self.criterion(torch.nn.functional.softmax(new_out_puts_batch, dim=1), self.init_y[batch_indx])
+            loss = self.criterion(new_out_puts_batch, self.init_y[batch_indx])
             batch_num = len(batch_indx)
             bias_parameters_grads = torch.autograd.grad(loss.sum(), new_out_puts_batch, retain_graph=True)[0]
 

deepcore/methods/gradmatch.py

@@ -142,7 +142,7 @@ def calc_gradient(self, index=None, val=False):
         for i, (input, targets) in enumerate(batch_loader):
             self.model_optimizer.zero_grad()
             outputs = self.model(input.to(self.args.device)).requires_grad_(True)
-            loss = self.criterion(torch.nn.functional.softmax(outputs, dim=1), targets.to(self.args.device)).sum()
+            loss = self.criterion(outputs, targets.to(self.args.device)).sum()
             batch_num = targets.shape[0]
             with torch.no_grad():
                 bias_parameters_grads = torch.autograd.grad(loss, outputs, retain_graph=True)[0].cpu()

deepcore/methods/grand.py

@@ -38,7 +38,7 @@ def finish_run(self):
         for i, (input, targets) in enumerate(batch_loader):
             self.model_optimizer.zero_grad()
             outputs = self.model(input.to(self.args.device))
-            loss = self.criterion(torch.nn.functional.softmax(outputs.requires_grad_(True), dim=1),
+            loss = self.criterion(outputs.requires_grad_(True),
                                   targets.to(self.args.device)).sum()
             batch_num = targets.shape[0]
             with torch.no_grad():
@@ -60,7 +60,7 @@ def select(self, **kwargs):
 
         for self.cur_repeat in range(self.repeat):
             self.run()
-            self.random_seed = int(time.time() * 1000) % 100000
+            self.random_seed = self.random_seed + 5
 
         self.norm_mean = torch.mean(self.norm_matrix, dim=1).cpu().detach().numpy()
         if not self.balance:

deepcore/methods/submodular.py

@@ -62,7 +62,7 @@ def calc_gradient(self, index=None):
         for i, (input, targets) in enumerate(batch_loader):
             self.model_optimizer.zero_grad()
             outputs = self.model(input.to(self.args.device))
-            loss = self.criterion(torch.nn.functional.softmax(outputs.requires_grad_(True), dim=1),
+            loss = self.criterion(outputs.requires_grad_(True),
                                   targets.to(self.args.device)).sum()
             batch_num = targets.shape[0]
             with torch.no_grad():