Subject: Marine Geospatial Ecology Tools (MGET) help
Text archives
| From: | "Jason Roberts" <> |
|---|---|
| To: | <> |
| Cc: | <> |
| Subject: | RE: [mget-help] negative binomial |
| Date: | Fri, 18 Dec 2009 17:21:50 -0500 |
Hi Mara,
I have not used a negative binomial before, but I took a look at the
documentation. I think it would not be too much trouble to implement the
mgcv "negbin" family for MGET 0.8.
I have one question that you may be able to answer, as a prior user of this
family. For negbin to work from the MGET Fit GAM tool, it looks like we need
to add two parameters:
1. Theta - this appears to be required by negbin. There appear to be several
ways of specifying it (e.g. a single value, a range of values, etc.).
Because if this, I think it would be simplest for MGET to accept Theta as a
string that it then uses to construct the call to R's negbin() function.
MGET would not perform any validation of Theta, other than to verify that
the user supplied something when negbin is selected as the model family. It
would be up to the user to ensure that he or she is providing a valid value,
and if it is invalid, it would be up to negbin() itself to issue a
reasonable error message.
2. Optimizer - it appears that it might be important to be able to specify
which optimizer is used by gam() when using the negbin family. I would add
this as an optional parameter. It is probably also useful for certain
problems when not using negbin, although nobody has requested it.
Does this approach sound right?
Happy holidays,
Jason
-----Original Message-----
From:
[mailto:]
Sent: Friday, December 18, 2009 8:30 AM
To:
Subject: [mget-help] negative binomial
Dear Jason and MGET team,
thanks for all the improvements you already did.
I was wondering whether it might be possibe to implement the "negative
binomial" family into the GAM options for the mgcv package (negbin(mgcv)).
During my research I realised that count data are often better modeled using
this family than by (quasi)poisson. Anyway, I also know that computation can
be
quite heavy and depending on the input the analysis might fail to converge.
Thanks a lot, merry christmas
Mara
I have not used a negative binomial before, but I took a look at the
documentation. I think it would not be too much trouble to implement the
mgcv "negbin" family for MGET 0.8.
I have one question that you may be able to answer, as a prior user of this
family. For negbin to work from the MGET Fit GAM tool, it looks like we need
to add two parameters:
1. Theta - this appears to be required by negbin. There appear to be several
ways of specifying it (e.g. a single value, a range of values, etc.).
Because if this, I think it would be simplest for MGET to accept Theta as a
string that it then uses to construct the call to R's negbin() function.
MGET would not perform any validation of Theta, other than to verify that
the user supplied something when negbin is selected as the model family. It
would be up to the user to ensure that he or she is providing a valid value,
and if it is invalid, it would be up to negbin() itself to issue a
reasonable error message.
2. Optimizer - it appears that it might be important to be able to specify
which optimizer is used by gam() when using the negbin family. I would add
this as an optional parameter. It is probably also useful for certain
problems when not using negbin, although nobody has requested it.
Does this approach sound right?
Happy holidays,
Jason
-----Original Message-----
From:
[mailto:]
Sent: Friday, December 18, 2009 8:30 AM
To:
Subject: [mget-help] negative binomial
Dear Jason and MGET team,
thanks for all the improvements you already did.
I was wondering whether it might be possibe to implement the "negative
binomial" family into the GAM options for the mgcv package (negbin(mgcv)).
During my research I realised that count data are often better modeled using
this family than by (quasi)poisson. Anyway, I also know that computation can
be
quite heavy and depending on the input the analysis might fail to converge.
Thanks a lot, merry christmas
Mara
Archives powered by MHonArc.