[FLR-list] Use of FLQuantPoint

[Mark Payne]

Thanks for the comments and suggestions. I think I'm starting to see a way forward. It sounds like the general advice would be to base everything on accessor methods, and so long as everything is (somehow) accessible, then its fine. However, there are some limitations with, the ability of current classes to store  uncertainties that I need to live with (I'm not so excited about the idea of developing FLStock_with_uncertainties!), and therefore there can't be a one size fits all solution. 

How does this sound for a set of methods / solutions?

1. A set of standard accessor functions, that return simple (error-free) FLQuants in the same way as for other classes e.g
stock.n()
harvest()
ssb()
fbar()
tsb()
rec()

2. Having all the estimated parameters, with uncertainties, readily accessible through an FLPar object (?), so that any other object or method that wants access to them (e.g. a plot(FLSAM) method) can have at them. I will store the estimate and standard deviation only, so that people can go further with that if they like...

Mark


________________________________________
Fra: flr-list-bounces at r-forge.wu-wien.ac.at [flr-list-bounces at r-forge.wu-wien.ac.at] På vegne af Laurie [lauriekell at googlemail.com]
Sendt: 10. oktober 2011 09:07
Til: flr-list at flr-project.org
Emne: Re: [FLR-list] Use of FLQuantPoint

Looking at all the data.frame slots, these currently store value, std and up.bnd & low.bnd

library(FLSAM)
data(NSH.sam)


but it appears that a normal distribution is assumed when calculating the bnds, and that these are the 95th percentiles. However, depending on
the application other percentiles might be more appropriate, but you can always calculate these if you know the mean & std & assumed distribution.

with(ssb(   NSH.sam), (up.bnd-value)/std)
with(logssb(NSH.sam), (up.bnd-value)/std)
with(ssb(   NSH.sam), (value-low.bnd)/std)
with(logssb(NSH.sam), (value-low.bnd)/std)

So returning the bounds is not necessary and the bounds being returned might not be the correct ones. Therefore it would be better to just return the mean & std and
have a method to calculate any required statistic. There is also redundancy as you return logssb as well as ssb (same for tsb & fbar etc).

Also FLQuantPoint was designed to summarise emprical distributions, but ssb etc come from an assumed distribution.

Also if you know wt, mat stock.n & harvest then you know ssb fbar etc. So agian there is redundancy.

How does fbar relate to harvest, presumably each F has a std & variance and fbar is function of these.

Laurie



On 10/08/2011 01:55 PM, Mark Payne wrote:

Ok, that makes sense, but I'm still a little unsure about how this should be implemented - as I understand it you're talking mainly about how to store the data, rather than pass it around. I currently have an array of accessor functions:

ssb(sam.object)
tsb(sam.object)
fbar(sam.object)
stock.n(sam.object)
harvest(sam.object)
rec(sam.object)

What sort of objects should these return, given that these quantites now can all have confidence intervals associated with them?

Mark

________________________________________
Fra: flr-list-bounces at r-forge.wu-wien.ac.at<mailto:flr-list-bounces at r-forge.wu-wien.ac.at> [flr-list-bounces at r-forge.wu-wien.ac.at<mailto:flr-list-bounces at r-forge.wu-wien.ac.at>] P&#229; vegne af Laurie [lauriekell at googlemail.com<mailto:lauriekell at googlemail.com>]
Sendt: 8. oktober 2011 12:18
Til: flr-list at flr-project.org<mailto:flr-list at flr-project.org>
Emne: Re: [FLR-list] Use of FLQuantPoint

On 10/08/2011 11:47 AM, Mark Payne wrote:


Hi,

I am a bit confused about the intended use of FLQuantPoints. The FLSAM assessment model returns the confidence intervals for nearly everything that it calculates, including, for example, the ssb. I am trying to write an accessor function that will return the ssb as an FLQuantPoint object. The problem is that the confidence intervals for this value are asymmetric, so it's not sufficient just to return the mean and the variance. I was wondering what the intention for the uppq and lowq dimension was? Is this the appropriate place to store the estimated confidence bounds? Or are they solely for cases where you have a large distribution that you are trying to characteriste non-parametrically ie through median and the quartiles...

Second question - is there a plural class, FLQuantPoints? I can't seem to find one at the moment....

Mark



You say ssb has an asymmetric distribution but that only matters  if it
is from an empirical distribution, e.g. derived from an MC simulation.
When I had a quick look at ssb, the upper & lower CIs were just +-1.96
times the CV. I.e. if you know the distribution and parameters you can
derive the CIs.
Also for what percentiles do you want the CIs? this will vary on a
case-by-case basis.

FLQuantPoint was designed to summarise the 6th dim (iter) to reduce
storage space. If you are not running MC simulations then there is no
need for FLQuantPoint.

Instead if you return the parameters and hessian as FLPar objects then
you can design methods to calculate quantities of interest.

Laurie


_______________________________________________
flr-list mailing list
flr-list at flr-project.org<mailto:flr-list at flr-project.org>
https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/flr-list
_______________________________________________
flr-list mailing list
flr-list at flr-project.org<mailto:flr-list at flr-project.org>
https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/flr-list