Merge pull request #241 from PaddlePaddle/fix_tensor

Fix zero dimension tensor issue

Author: Noisyntrain

apps/drug_drug_synergy/RGCN/train.py

@@ -87,7 +87,7 @@ def train(num_subgraph, graph, label_idx, epochs, sub_neighbours=[10, 10], init=
                 fpr, tpr, _ = roc_curve(y_true=ground_truth, y_score=pred_prob)
                 auc_v = auc(fpr, tpr)
                 print("sub_graph index : {} | epoch: {} | training loss: {:.4f} | AUC: {:.3f}".format(
-                sub_g, epoch, train_loss.numpy()[0], auc_v))
+                sub_g, epoch, float(train_loss), auc_v))
 
     return model
 

apps/drug_target_interaction/batchdta/pairwise/DeepDTA/utils.py

@@ -312,7 +312,7 @@ def model_eval(model,val_dataloader):
 
         for i_target_score in range(batch_smiles.shape[0]):
 
-            i_target_len = int(batch_len[i_target_score].numpy()[0])
+            i_target_len = int(batch_len[i_target_score])
             smiles = batch_smiles[i_target_score][0:i_target_len]
             target = batch_protein[i_target_score][0:i_target_len]
             y_label = batch_y[i_target_score][0:i_target_len].numpy()

apps/drug_target_interaction/batchdta/pairwise/GraphDTA/run_pairwise_GraphDTA_CV.py

@@ -195,9 +195,9 @@ def model_eval(model,val_dataloader,device):
                 i_data = i_data.to(device)                    
                 pred_scores = model.forward_single(i_data)
                 # get the predicted labels
-                i_target_pred_scores.append(pred_scores.cpu().numpy()[0])              
+                i_target_pred_scores.append(float(pred_scores))              
                 # get the true labels
-                i_target_y_label.append(i_data.y.cpu().numpy()[0])
+                i_target_y_label.append(float(i_data.y.cpu()))
 
             i_target_pred_scores = np.array(i_target_pred_scores)
             i_target_y_label = np.array(i_target_y_label)

apps/drug_target_interaction/batchdta/pairwise/Moltrans/helper/utils/paddle_tensor.py

@@ -32,7 +32,7 @@ def item(self):
     """
     Item function
     """
-    return self.numpy()[0]
+    return float(self)
 
 
 @add_tensor_function

apps/drug_target_interaction/batchdta/pairwise/Moltrans/run_pairwise_Moltrans_CV.py

@@ -297,7 +297,7 @@ def model_eval(model,val_dataloader,len_SMILES,len_target):
 
         for i_target_score in range(batch_x.shape[0]):
 
-            i_target_len = int(batch_len[i_target_score].numpy()[0])
+            i_target_len = int(batch_len[i_target_score])
             smiles = batch_x_smiles[i_target_score][0:i_target_len]
             target = batch_x_protein[i_target_score][0:i_target_len]
             smiles_mask = batch_x_smiles_mask[i_target_score][0:i_target_len]

apps/drug_target_interaction/batchdta/pairwise/Moltrans/run_pairwise_Moltrans_bindingDB.py

@@ -282,7 +282,7 @@ def model_eval(model,val_dataloader,len_SMILES,len_target):
 
         for i_target_score in range(batch_x.shape[0]):
 
-            i_target_len = int(batch_len[i_target_score].numpy()[0])
+            i_target_len = int(batch_len[i_target_score])
             smiles = batch_x_smiles[i_target_score][0:i_target_len]
             target = batch_x_protein[i_target_score][0:i_target_len]
             smiles_mask = batch_x_smiles_mask[i_target_score][0:i_target_len]

apps/drug_target_interaction/batchdta/pointwise/DeepDTA/train_bindingdb.py

@@ -60,7 +60,7 @@ def training(model, training_loader, optim):
         optim.clear_grad()
         loss.backward()
         optim.step()
-        res_loss = loss.numpy()[0]
+        res_loss = float(loss)
     return res_loss
 
 

apps/drug_target_interaction/batchdta/pointwise/DeepDTA/train_davis.py

@@ -60,7 +60,7 @@ def training(model, training_loader, optim):
         optim.clear_grad()
         loss.backward()
         optim.step()
-        res_loss = loss.numpy()[0]
+        res_loss = float(loss)
     return res_loss
 
 

apps/drug_target_interaction/batchdta/pointwise/DeepDTA/train_kiba.py

@@ -63,7 +63,7 @@ def training(model, training_loader, optim):
         optim.clear_grad()
         loss.backward()
         optim.step()
-        res_loss = loss.numpy()[0]
+        res_loss = float(loss.numpy())
     return res_loss
 
 

apps/drug_target_interaction/batchdta/pointwise/Moltrans/helper/utils/paddle_tensor.py

@@ -32,7 +32,7 @@ def item(self):
     """
     Item function
     """
-    return self.numpy()[0]
+    return float(self)
 
 
 @add_tensor_function

apps/drug_target_interaction/moltrans_dti/helper/utils/paddle_tensor.py

@@ -32,7 +32,7 @@ def item(self):
     """
     Item function
     """
-    return self.numpy()[0]
+    return float(self.numpy())
 
 
 @add_tensor_function

apps/fewshot_molecular_property/chem_lib/models/trainer.py

@@ -294,7 +294,7 @@ def train_step(self):
             losses_eval.backward()
             self.optimizer.step()
 
-            print('Train Epoch:',self.train_epoch,', train update step:', k, ', loss_eval:', losses_eval.numpy()[0])
+            print('Train Epoch:',self.train_epoch,', train update step:', k, ', loss_eval:', float(losses_eval))
 
         return self.model.layers
 

apps/molecular_generation/SD_VAE/train_zinc.py

@@ -122,9 +122,9 @@ def _train_epoch(model, data_loader, epoch, kl_weight, optimizer=None):
             optimizer.clear_grad()
 
         # Log
-        kl_loss_values.append(kl_loss.numpy()[0])
-        perplexity_loss_values.append(perplexity.numpy()[0])
-        loss_values.append(loss.numpy()[0])
+        kl_loss_values.append(float(kl_loss))
+        perplexity_loss_values.append(float(perplexity))
+        loss_values.append(float(loss))
         lr = (optimizer.get_lr()
                   if optimizer is not None
                   else 0)

apps/pretrained_compound/ChemRL/GEM-2/src/paddle_utils.py

@@ -37,8 +37,8 @@ def dist_mean(array, distributed=False):
     n = len(array)
     x_sum = 0 if n == 0 else np.sum(array)
     if distributed:
-        n = dist_all_reduce(paddle.to_tensor(n, dtype='int64')).numpy()[0]
-        x_sum = dist_all_reduce(paddle.to_tensor(x_sum, dtype='float32')).numpy()[0]
+        n = int(dist_all_reduce(paddle.to_tensor(n, dtype='int64')))
+        x_sum = float(dist_all_reduce(paddle.to_tensor(x_sum, dtype='float32')))
     x_mean = 0 if n == 0 else x_sum / n
     return x_mean
 
@@ -47,14 +47,14 @@ def dist_sum(array, distributed=False):
     n = len(array)
     x_sum = 0 if n == 0 else np.sum(array)
     if distributed:
-        x_sum = dist_all_reduce(paddle.to_tensor(x_sum, dtype='float32')).numpy()[0]
+        x_sum = float(dist_all_reduce(paddle.to_tensor(x_sum, dtype='float32')))
     return x_sum
 
 
 def dist_length(array, distributed=False):
     n = len(array)
     if distributed:
-        n = dist_all_reduce(paddle.to_tensor(n, dtype='int64')).numpy()[0]
+        n = int(dist_all_reduce(paddle.to_tensor(n, dtype='int64')))
     return n
 
 

apps/pretrained_compound/ChemRL/GEM-2/train_gem2.py

@@ -80,7 +80,7 @@ def get_train_steps_per_epoch(dataset_len, args):
         min_data_len = paddle.to_tensor(dataset_len)
         from paddle.distributed import ReduceOp
         dist.all_reduce(min_data_len, ReduceOp.MIN)
-        dataset_len = min_data_len.numpy()[0]
+        dataset_len = int(min_data_len)
         logging.info(f'min dataset len: {dataset_len}')
     return int(dataset_len / args.batch_size) - 5
 

apps/protein_folding/helixfold-single/tape/others/utils.py

@@ -48,8 +48,8 @@ def dist_all_reduce(x, return_num=False, distributed=False):
     n = len(x)
     x_sum = 0 if n == 0 else np.sum(x)
     if distributed:
-        n = dist.all_reduce(paddle.to_tensor(n, dtype='int64')).numpy()[0]
-        x_sum = dist.all_reduce(paddle.to_tensor(x_sum, dtype='float32')).numpy()[0]
+        n = int(dist.all_reduce(paddle.to_tensor(n, dtype='int64')))
+        x_sum = float(dist.all_reduce(paddle.to_tensor(x_sum, dtype='float32')))
     x_mean = 0 if n == 0 else x_sum / n
     if return_num:
         return x_mean, n
@@ -62,8 +62,8 @@ def dist_mean(x, distributed=False):
     n = len(x)
     x_sum = 0 if n == 0 else np.sum(x)
     if distributed:
-        n = dist.all_reduce(paddle.to_tensor(n, dtype='int64')).numpy()[0]
-        x_sum = dist.all_reduce(paddle.to_tensor(x_sum, dtype='float32')).numpy()[0]
+        n = int(dist.all_reduce(paddle.to_tensor(n, dtype='int64')))
+        x_sum = float(dist.all_reduce(paddle.to_tensor(x_sum, dtype='float32')))
     x_mean = 0 if n == 0 else x_sum / n
     return x_mean
 
@@ -73,15 +73,15 @@ def dist_sum(x, distributed=False):
     n = len(x)
     x_sum = 0 if n == 0 else np.sum(x)
     if distributed:
-        x_sum = dist.all_reduce(paddle.to_tensor(x_sum, dtype='float32')).numpy()[0]
+        x_sum = float(dist.all_reduce(paddle.to_tensor(x_sum, dtype='float32')))
     return x_sum
 
 
 def dist_length(x, distributed=False):
     """tbd"""
     n = len(x)
     if distributed:
-        n = dist.all_reduce(paddle.to_tensor(n, dtype='int64')).numpy()[0]
+        n = int(dist.all_reduce(paddle.to_tensor(n, dtype='int64')))
     return n
 
 

apps/protein_folding/helixfold/utils/metric.py

@@ -30,8 +30,8 @@ def dist_all_reduce(x, return_num=False, distributed=False):
     x_num = len(x)
     x_sum = 0 if x_num == 0 else np.sum(x)
     if distributed:
-        x_num = dp.all_reduce(paddle.to_tensor(x_num, dtype='int64')).numpy()[0]
-        x_sum = dp.all_reduce(paddle.to_tensor(x_sum, dtype='float32')).numpy()[0]
+        x_num = int(dp.all_reduce(paddle.to_tensor(x_num, dtype='int64')))
+        x_sum = float(dp.all_reduce(paddle.to_tensor(x_sum, dtype='float32')))
     x_mean = 0 if x_num == 0 else x_sum / x_num
     if return_num:
         return x_mean, x_num