DSL-Lab
The problem of the logits'shape in the train_epoch function in the dgl_main.py
Hi, i am very excited for your work, so i am running your codes. there is a problem for me in the train_epoch function in the dgl_main.py.
Actually, in the train_epoch, there is a parameter named as "logits", the computation of logits as follow
logits = model(e, u, g, length)
so logits'type is tuple, so there will be a error such as "'tuple' object has no attribute 'to'".
Now i solve this problem by chatgpt
#chatgpt if isinstance(logits, tuple): logits = logits[0] #chatgpt # Adjust logits and y shapes if necessary logits = logits.squeeze(dim=-1) # 去除多余的维度
By adding these codes, the program can run.
However, logits is a tuple, the codelogits = logits[0] just gains the first element of logits, is this ture?
I will be happy for your reply!
Best wishes!
