In Subgradient Descent, I didn't quite understand the statement "The complete objective for SVM is a sum, not an average as in earlier SGD examples! Be careful about the constant N (size) term!". What should I do with the N term?
The objective being a sum or an average in general does not matter much because multiplying by a positive number does not change an optimization problem (does not change the minimizer which is what is of interest mostly). However, you should pay attention to the regularization parameter as it scale will change.
Lab 7 Question
In Subgradient Descent, I didn't quite understand the statement "The complete objective for SVM is a sum, not an average as in earlier SGD examples! Be careful about the constant N (size) term!". What should I do with the N term?
1
The objective being a sum or an average in general does not matter much because multiplying by a positive number does not change an optimization problem (does not change the minimizer which is what is of interest mostly). However, you should pay attention to the regularization parameter as it scale will change.
Add comment