diff --git a/examples/scsn_polarity_lstm_training.py b/examples/scsn_polarity_lstm_training.py
index 485c41d..d943a10 100644
--- a/examples/scsn_polarity_lstm_training.py
+++ b/examples/scsn_polarity_lstm_training.py
@@ -33,7 +33,7 @@
     # Prepare dataset
     dsets.set_memory_limit(10 * 1024 ** 3) # first number is GB
     # dset = dsets.Wenchuan(path='/home/qszhai/temp_project/deep_learning_course_project/cpic', download=False,sample_transform=waveform_transform)
-    dset = dsets.Wenchuan(path='/home/qszhai/temp_project/deep_learning_course_project/first_motion_polarity/scsn_data/train_npy', download=False, sample_transform=waveform_transform)
+    dset = dsets.SCSN_polarity(path='/home/qszhai/temp_project/deep_learning_course_project/first_motion_polarity/scsn_data/train_npy', download=False, sample_transform=waveform_transform)
 
     # Split datasets into training and validation
     train_length = int(len(dset) * 0.8)
@@ -45,7 +45,8 @@
     val_loader = DataLoader(val_set, batch_size=10000, shuffle=False, num_workers=4)
 
     # Prepare trainer
-    #trainer = Trainer(cpic(), CrossEntropyLoss(), lr=0.1)
+    # trainer = Trainer(cpic(), CrossEntropyLoss(), lr=0.1)
+    # note: please use only 1 gpu to run LSTM, https://github.com/pytorch/pytorch/issues/21108
     model_conf = {"hidden_size": 64}
     plt = polarity(**model_conf)
     trainer = Trainer(plt, CrossEntropyLoss(), lr=0.001)