Questions about logistic regression

  1. How should my logistic regression results (e.g. odds ratio, p values and R^2) be reported on scientific papers?
    If i were to apply 10 fold cross validation, i’d get 10 sets of varying results. Which should i report?

  2. Is it possible to optimize lambda in a similar fashion to the way gradient descent is performed? (Plot average R^2 values from cross vals against lambda)

Hello @Isaac_Lam, welcome to our community!

  1. I suggest you to read at least the summary of this Wikipedia article on the purpose of cross validation and how you may deal with the 10 results.

  2. What is your purpose of doing cross-validation?

  1. Cross validation is a technique for you to choose a good value for your hyperparameter lambda.

  2. What does your plot look like?

  3. Gradient descent tunes weights to reduce cost. Lambda (regularization) is known to add cost, so if you let gradient descent tunes your lambda, it will only make it zero.


… which leads to overfitting the training set. Which is a bad thing.