update experiment result in GaAN example

examples/GaAN/README.md

@@ -13,7 +13,13 @@ The ogbn-proteins dataset will be downloaded in directory ./dataset automaticall
 ## How to run
 ```bash
 python train.py --lr 1e-2 --rc 0 --batch_size 1024 --epochs 100
-``` 
+```
+
+or
+```bash
+source main.sh
+```
+
 ### Hyperparameters
 - use_gpu: whether to use gpu or not
 - mini_data: use a small dataset to test code
@@ -32,4 +38,4 @@ python train.py --lr 1e-2 --rc 0 --batch_size 1024 --epochs 100
 We train our models for 100 epochs and report the **rocauc** on the test dataset.
 |dataset|mean|std|
 |-|-|-|
-|ogbn-proteins|0.7786|0.0048|
+|ogbn-proteins|0.7803|0.0073|

examples/GaAN/main.sh

-python3 train.py --epochs 100 --lr 1e-2 --rc 0 --batch_size 1024 --gpu_id 4 --exp_id 39
\ No newline at end of file
+python3 train.py --epochs 100 --lr 1e-2 --rc 0 --batch_size 1024 --gpu_id 0 --exp_id 0
\ No newline at end of file

examples/GaAN/train.py

@@ -194,16 +194,12 @@ if __name__ == "__main__":
             print("Update: new {}, old {}".format(valid_rocauc, best_valid))
             best_valid = valid_rocauc
             
-#             fluid.io.save_vars(executor=exe, dirname='./vars/'+str(args.exp_id), main_program=val_program,
-#                               predicate=lambda x: True)
             fluid.io.save_params(executor=exe, dirname='./params/'+str(args.exp_id), main_program=val_program)
             
 
     print("Test Stage".center(50, "="))
     log_text.add_record(args.epochs+1, "Test Stage".center(50, "="))
     
-#     fluid.io.load_vars(executor=exe, dirname='./params/'+str(args.exp_id), main_program=val_program,
-#                       predicate=lambda x: True)
     fluid.io.load_params(executor=exe, dirname='./params/'+str(args.exp_id), main_program=val_program)
     
     test_loss, test_rocauc = valid_epoch(

pgl/graph.py

@@ -593,7 +593,7 @@ class Graph(object):
             edges = self._edges[eid]
         else:
             edges = np.array(edges, dtype="int64")
-
+            
         sub_edges = graph_kernel.map_edges(
             np.arange(
                 len(edges), dtype="int64"), edges, reindex)