I am working on an assignment for neural networks and I need to predict an angle. firstly the last layer of the neural network directly predicts the angle, secondly, the last layer gives as an output the sin and cos of this angle, and the final prediction of the angle are given from arctan(sin/cos).

Why is the second way for predicting the angle is better and will it be efficient to predict the only sin and predict the angle with arcsin(sin)?

  • $\begingroup$ Presumably because it is easier to define a circular loss (teaching the model that 0° = 360°) with sin and cos $\endgroup$
    – oW_
    Mar 15 at 21:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.