search

reichlab / 2017-2018-cdc-flu-contest

Code and submissions for 2017-2018 CDC flu prediction contest
3 stars 2 forks source link

KDE Probability of no onset being calculated across all regions #17

Closed nickreich closed 7 years ago

nickreich commented 7 years ago

Not sure if this is intentional, but pretty sure that the code at https://github.com/reichlab/2017-2018-cdc-flu-contest/blob/c06d0180a86f30b43d0f87d5d4e9c77bf7a4a267/R/kde-utils.R#L134 is calculating, across all regions, the probability of having no onset, when this is fitting a model for a single region. @elray1 was this an intentional decision?

nickreich commented 7 years ago

If not intentional, then we can should the loop over regions in kde-utils.R.

elray1 commented 7 years ago

This is intentional, some regions had no seasons without onset in training, but since there were few training seasons an estimate of zero probability for a given region is unwarranted. The idea was that pooling regions would give a larger sample size for estimating probability of no onset.

nickreich commented 7 years ago

Thought that might be the case. Thanks for confirming the intent!

On 22 Jun 2017, at 14:37, elray1 wrote:

This is intentional, some regions had no seasons without onset in training, but since there were few training seasons an estimate of zero probability for a given region is unwarranted. The idea was that pooling regions would give a larger sample size for estimating probability of no onset.

-- You are receiving this because you authored the thread. Reply to this email directly or view it on GitHub: https://github.com/reichlab/2017-2018-cdc-flu-contest/issues/17#issuecomment-310466055