How does it work for me to call an object in a function that doesnt have it as a parameter in python/pytorch?

Im learning neural network in PyTorch and i came across with that:

#Loss function
criterion = nn.MSELoss()

#Optimizer
from torch import optim    
optimizer = optim.Adam(MLP.parameters(), lr=args['lr'], weight_decay=args['weight_decay'])


def train(train_loader, MLP, epoch): #MLP is the model
    
    MLP.train()
    start = time.time()
    
    epoch_loss = []

    for batch in train_loader:
        
        sample, label = batch
        
        optimizer.zero_grad()

        #Forward
        pred = MLP(sample)
        loss = criterion(pred, label)
        epoch_loss.append(loss.data)

        #Backward
        loss.backward()
        optimizer.step()

    epoch_loss = np.asarray(epoch_loss)
    
    end = time.time()
    print('Epoch: {}, Loss: {:.4f} +/- {:.4f}, Time: {}'.format(epoch+1, epoch_loss.mean(), epoch_loss.std(), end-start))
    
    return epoch_loss.mean()

Well, "criterion" and "optimizer" are objects that i didnt pass as parameters for my function "train" like i did with the model (MLP), but it worked. Does it work for any function or is it just a PyTorch’s thing?

>Solution :

This is not a Pytorch thing, these are called global (as opposed to local) variables. I would advise to get more familiar with the Python language and programming in general, if you want to get a grip on Pytorch.