Including code for training from checkpoint

[imjohnzakkam]

Updated the simclr.py file with including code for loading from checkpoints and other minor changes

[Yadino]

in load_checkpoint you start by doing a check
if(os.path.exists(filepath)):

Yet you try to catch an exception when you call this function in simclr.py. It will not throw an exception because of this check, but rather fail quietly.

Better do something like

def load_checkpoint(model, filepath):
try:
assert os.path.exists(filepath)
ckpt = torch.load(filepath)
model.load_state_dict(ckpt['state_dict'])
epoch = ckpt['epoch']
return model, epoch
except Exception:
raise InvalidCheckpointPath()