Ad
  • Custom User Avatar

    There is only 1 toy data set and 1 big data set used for the kata. The tests are hence laughably weak (not to mention lack of random tests).

  • Custom User Avatar
    main.rb:344:in `<main>': undefined method `last_j_differential' for #<LinearRegression:0x0000555ad184e500> (NoMethodError)
    
  • Custom User Avatar

    As the kata currently stands it's univariate linear regression + cost function, which is a duplicate of the two katas linked in the starting paragraph in the description. Solutions can be directly copied to this kata and solve the related part.

  • Custom User Avatar

    Univariate linear regression do not even need gradiant descent in the first place, so I don't see the point of the kata above the usual linear regression, besides kata author wanting to dump the entire lecture they encountered because they thought they learnt everything related to linear regression from it. Notably univariate linear regression has a very efficient and numerically stable closed-form solution. Gradient descent is only advantagous in multivariate cases.

    In any case, this point and the rebuttal on the lecture's usefulness has been detailed in https://stats.stackexchange.com/a/278794

  • Custom User Avatar

    Why is input in the form [1, x] when it's just one variable? This is very unnatural, and unexplained in the kata description.