Github user actuaryzhang commented on the issue:
https://github.com/apache/spark/pull/16344
@yanboliang Thanks for the feedback. However, I'm not sure why we need to
be consistent with R on this one. The usage of 'tweedie' glm almost always uses
`link.power = 0, 1, -1`. There is barely any use case for the default
`1-variancePower` link. I think a better design is to set the default to the
one that is used more often, and make it clear that `we only support identity,
log, inverse and sqrt link functions`.
On the other hand, even if we want to be consistent with R, I actually
wrote the R package `cplm` which fits the tweedie
[glm](http://finzi.psych.upenn.edu/library/cplm/html/cpglm.html) as well as
[mixed effect
model](https://www.rdocumentation.org/packages/cplm/versions/0.7-4/topics/cpglmm).
In that package I set the default to the `log` link and never had an issue.
I don't think I will implement `linkPower`, but I'm fine if you want to do
it. However, if the lack of `linkPower` means this PR won't get merged, please
let me know and we can close this and move on.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
