Tom Wenseleers

Many thanks for this - that's very helpful! I was just sort of hoping that allowing the observation weights to be a matrix instead of a vector might be a...

Ha that's a shame... I should say that even for the n > p case I mostly found glmnet to outperform glum, though I haven't tried with super huge n...

Thanks for that! FYI - I was benchmarking the speed of some GLM fitting engines in R the other day (speedglm, fastglm, parglm, bigglm, glmnet's bigGlm), and for tall datasets...

Ha that's great - yes please send it over (you can send it to [email protected])! My scam models also use the weights argument btw to take into account unequal variances...

Just realised that the scam::predict and mgcv:predict functions also return the SE on the predictions based on the posterior distribution of the parameters if you set se.fit=TRUE - maybe the...

Was wondering if the inbuilt scam or gam se.fit=TRUE in predict together with using the bisection method for the inverse estimate wouldn't be good enough though? Something like ``` model.fit...

Ha sorry I see now that confidence intervals were also still on the TO DO list... :-) I was indeed thinking that the delta method could be used for that......

Yes would be very useful - packages like glmnet and abess do support sparse covariate matrices, but would be good to have that in ncvreg too. If the package would...

Do you think it might be possible to implement this? I'm still looking for this... Right now the only package with MCP or SCAD penalties with support for box constraints...