Why does accuracy need argmax and not loss function when there are multiple class output probabilities?

[sohumgautam17]

Why does accuracy get argmax and loss not? This is in the computer vision section of the course @ 16:48 but its also throughout that section. Heres the code snippet:

y_eval = model(X)
loss+=loss_fn(y_eval, y)
acc = accuracy_fn(y, y_eval.argmax(dim=1))

This is the rest of the training loop
"""
torch.manual_seed(42)
def evalmodel(model:torch.nn.Module, dataloader:torch.utils.data, loss_fn:torch.nn.Module, accuracy_fn):
loss, acc = 0, 0
torch.eval()

with torch.inference_mode():
for X, y in dataloader:

  y_eval = model(X)
  loss+=loss_fn(y_eval, y)
  acc = accuracy_fn(y, y_eval.argmax(dim=1))


loss /= len(dataloader) # get the average
acc /= len(dataloader)


return{"model_name": model.__class__.__name__,
       "model_Loss": loss.item(), 
       "model_acc" : acc}

"""

[LuluW8071]

In background,

• A loss function measures the difference between the actual values and the predictions made by model.
  Example:

  5-3=2  # where '5' is actual and `3` is predicted value and `2` is distance between them (called loss)

• Accuracy is calculated by comparing the predicted labels against the actual labels. Specifically for each prediction, it checks if index of the maxm. value in predicted probabilities corresponds to the index of the actual label.
  Example:
  Suppose [0.1, 0.7, 0.2] for three classes generated by training a model.

  0.1 # index0
  0.7 # index1
  0.2 # index2

  The index of the maximum probability(0.7) is 1.

  acc = accuracy_fn(y, y_eval.argmax(dim=1)) # .argmax selects index of max probabality

if you do print statements like this you can see what goes on background

print(f'Loss calculation: {loss_fn(y_eval, y)}')
loss += loss_fn(y_eval, y)
print('Actual label:', y)
print('Predicted label:', y_eval.argmax(dim=1))
acc = accuracy_fn(y, y_eval.argmax(dim=1))