search

fhui28 / CBFM

Spatio-temporal joint species distribution modeling using community-level basis functions
13 stars 2 forks source link

More plotting #2

Closed fhui28 closed 1 year ago

fhui28 commented 3 years ago

termplot.CBFM for parametric plots; coefplot.CBFM; anything fancier like plot.gam for smooth functions?

fhui28 commented 2 years ago

A basic version would exploit using plot.gam off the final GAM fit in the PQL estimation algorithm. But can only do point estimates; see plot.gam helpful for more details.

Alternatively manual construction of the function is possible...but hard yards here!

chrishaak commented 2 years ago

Just some thoughts relevant to plotting; there is the "gratia" package from gavin simpson, and I just happened upon "mgcViz" that also exploits ggplot2 and looks to be quite useful... (and easily includes plots for parametric terms as well as smooths)

fhui28 commented 2 years ago

Starting example to look into

library(tidyverse)
library(gratia)
library(mgcv)
dat <- data_sim("eg1", n = 100, dist = "normal", scale = 2, seed = 2)

m1 <- gam(y ~ x0 + s(x1) + s(x2) + s(x3), data = dat, method = "REML")
draw(m1, parametric = TRUE)

parametric_effects(m1) 
smooth_estimates(m1)
fhui28 commented 2 years ago

Package now implemented employs parametric_effects and smooth_estimates, but SEs and CIs are not possible.

Will need to ask Gavin Simpson about gratia and how to return specific things within his function used for constructing the predictions. I will make a note of this on github, but please don't hold your breath!

fhui28 commented 2 years ago

Standard errors are implemented for both all_parametric_effects and all_smooth_estimates returned from a CBFM object. Should be correct I hope!

fhui28 commented 1 year ago

Should either make an option in main CBFM to turn this off and/or create a separate function to do this. I prefer the latter.

chrishaak commented 1 year ago

WRT to a separate function, are you thinking that it would operate on a fitted CBFM model object? i.e., one would run a model, and then could generate the plotting object at any point afterward? If so, then that seems like a nice approach...

On Sun, Nov 20, 2022 at 4:34 PM Francis KC Hui @.***> wrote:

Should either make an option in main CBFM to turn this off and/or create a separate function to do this. I prefer the latter.

— Reply to this email directly, view it on GitHub https://github.com/fhui28/CBFM/issues/2#issuecomment-1321248486, or unsubscribe https://github.com/notifications/unsubscribe-auth/AORLQXE6LKAT6AIWSHM4MPDWJKKMVANCNFSM5AO6Q3EQ . You are receiving this because you commented.Message ID: @.***>

fhui28 commented 1 year ago

Yes that's the plan! But do it after the MEE revision

fhui28 commented 1 year ago

Done and now in postMEE branch. Will merge into main branch after manuscript is formally accepted somewhere.