meaning of `args.batchsize`

[KrisNguyen135]

Hi, thanks for publishing the code; very cool work! Not really an issue but more of a question about the code...

I'm trying to understand what args.batchsize (https://github.com/ShengdingHu/GraphPolicyNetworkActiveLearning/blob/main/src/train.py#L29) does in the code. Does it, as the name suggests, denote how many queries you can make at each iteration of the active learning loop? Or is it the size of each batch when you train the policy network?

Maybe your response to the above will answer the following questions, but I'm putting them here:

• In player.py, each mask (for training, validation, and test) is a tensor of size args.batchsize-by-the number of nodes (https://github.com/ShengdingHu/GraphPolicyNetworkActiveLearning/blob/main/src/utils/player.py#L151). Why do you need args.batchsize in the first dimension there? It seems to me you would only need a 1d tensor of length equal to the number of nodes to keep track of which data points are in which set.
• What does getPool() in player.py (https://github.com/ShengdingHu/GraphPolicyNetworkActiveLearning/blob/main/src/utils/player.py#L76) do exactly? And how is it being used in https://github.com/ShengdingHu/GraphPolicyNetworkActiveLearning/blob/main/src/train.py#L129-L130?
• Why do we need to resize the output of each GCN layer to add a second dimension for args.batchsize (https://github.com/ShengdingHu/GraphPolicyNetworkActiveLearning/blob/main/src/utils/policynet.py#L41)?

Thanks in advance!