batch size?

[cwsmith]

It appears that we are specifying a batch size of 1 here:

reconClassifier/XPointMLTest.py

Lines 657 to 658 in 9d8d10b

	train_loader = DataLoader(train_dataset, batch_size=1, shuffle=False)
	val_loader = DataLoader(val_dataset, batch_size=1, shuffle=False)

Should we test other sizes? How can this affect the learning rate and/or performance?

[cwsmith]

A command line arg for batch size was added with: 36ea19b. The default is a batch size of one.

[jimlo7760]

We will try a larger batch size. Batch size is the number of frames that the model will be processed together before updating the weight (1 epoch). From what google research tuning book states, the batch size should be the largest possible number that hardware supports, which will help the model generalize better. https://github.com/google-research/tuning_playbook#choosing-the-batch-size