Join GitHub today
GitHub is home to over 36 million developers working together to host and review code, manage projects, and build software together.Sign up
robust covariance, cov_type in fit #1870
continuing after PR #1867
two possible problems using a standalone function, e.g.
some problems in the design:
Since we are changing some attributes in the results instance, we need to do it at the right time.
example NegativeBinomial nb uses log(alpha) internally but reports alpha. (This is different from models were the original parameterization is in transformed params, e.g. the link functions)
referenced this pull request
Aug 8, 2014
possible bug in code so far: how do we handle scale if it separately estimated.
In a new set of test I'm comparing more GLM models to equivalent other models.
Current test cases for MLE robust covariances have scale=1, Poisson, NegativeBinomial, and GLM also agrees with Logit.
Difference between estimating equations with a canceled multiplicative term versus score/score_obs which should be the derivative of the full loglikelihood function.
TODO: add score_obs and Hessian (correctly scaled) to RegressionModels, OLS,...
This was referenced
Aug 9, 2014
rebased and force pushed.
comment to earlier comment:
scale wasn't a problem in the code when comparing GLM to OLS,
Non-native scale will or might still be a problem in overdispersed Poisson, or when scale is fixed, but not here.
Aside: score and Hessian of OLS have a division by the scale/sigma**2.