beckyfisher / fssgam Goto Github PK

Hi,` Becky! I think this code is fantastic, but I'm having trouble getting an output for fit.model.set because I keep getting the following error: 'Error in y_dat[, 1] : incorrect number of dimensions'.

I have tried to follow your examples for constructing the initial model and generating the model set (both of which have successfully given list outputs), but I have had no luck in getting past model fitting. If you can see where I may be going wrong, any guidance would be greatly appreciated!

##### Define Model Variables #####

cont.preds = c("Dist.km", "Depth", "Bathy", "Noise.Level") # Use as continuous predictors.
cat.preds = c("Shipping.Lane", "LightRegime") # Use as categorical predictors.
null.vars = "Transit" # Use as random effect and null model.


##### Fit model set #####

Model1 = uGamm(PresAbs ~ 1,
                 random =~ (1|Transit), 
                 data = BW, 
                 family = binomial(link = "logit"),
                 lme4 = TRUE)
  
  model.set = generate.model.set(use.dat = BW, 
                                 max.predictors = 6,
                                 test.fit = Model1,
                                 k = 3,
                                 pred.vars.cont = cont.preds,
                                 pred.vars.fact = cat.preds)
  
  out.list = fit.model.set(model.set, max.models = 720)
  names(out.list)

Add some error catching around the null.term passed to generate.model.set

Running the code generate.model.set returns the above error message

@beckyfisher What advice can we give on when to report a top model or not?
For example - if a model explains <10% variance should it it even be reported in a table?

FSSgam currently doesn't support fitting via gamm to allow for a correlation structure because the null model contains no random effects.

When family is passed as an element of a vector, all models fail to fit when parallel is TRUE.

Hi, thanks for your code. I'm trying out this method in my analysis, however an error is coming up when I specify factor variables in full.subsets.gam()

This is the error:

Error in sort.list(y) : 'x' must be atomic for 'sort.list'
Have you called 'sort' on a list?

Seems to coming out from a sort function

The function runs if I don't specify factors.

There is an issue with fitting the full model set when the family nb() is used. Most models (but not all) appear to fail. Note the same issues occurs in the use of lapply

In addition to the relative importance of predictors, is it possible to get their P-values?

I'm trying to make a tutorial based on this data and can't seem to troubleshoot why I'm getting this error for the code following ### now fit the models ---------------------------------------------------------
Any assistance appreciated

Error in names(dat) <- object$term :
'names' attribute [1] must be the same length as the vector [0]

Does linear.vars need to be in pred.vars or not? Table 1 needs to be clear