What’s the meaning of `loss`

& `acc`

? I mean they are loss and accuracy of training data but why the sum is not equal to 1?

Accuracy is a method for measuring performance based on the actual value and the predicted value.

Loss is a performance measure that is based on how much the predicted value varies from the actual value.

As they are different performance measures, their sum is not 1 (in most cases).

`acc`

refers to how often your model predicted the correct class.

`loss`

refers to how close to the label your predictions were.

Consider a single binary classification where your prediction is 0.9 and your label is 1 (I’ll use mean absolute error as my loss function). Your prediction was closer to 1 than it was to 0, so it predicted correctly 1 out of 1 times, and your `acc`

is 1.0 (100%). Your `loss`

is |0.9 - 1| = 0.1, so you were close, but your model will use that loss to make its next prediction closer.

Got it, they are the respective values instead of being some probabilities or percentages obtained by simple subtraction.