predict for bsignal seems broken

[pfistfl]

The example from the vignette FLAM_fuel seems broken, the "predict" step does lead to errors

# From the vignette:
data(fuelSubset)
fuel <- fuelSubset
fuel$dUVVIS <- t(apply(fuel$UVVIS, 1, diff))
fuel$dNIR <- t(apply(fuel$NIR, 1, diff))
fuel$duvvis.lambda <- fuel$uvvis.lambda[-1]
fuel$dnir.lambda <- fuel$nir.lambda[-1]

modH2O <- FDboost(h2o ~ bsignal(UVVIS, uvvis.lambda, knots=40, df=4)
                    + bsignal(NIR, nir.lambda, knots=40, df=4)
                    + bsignal(dUVVIS, duvvis.lambda, knots=40, df=4)
                    + bsignal(dNIR, dnir.lambda, knots=40, df=4),
                    timeformula=~bols(1), data=fuel)

# Predict errors
predict(modH2O, fuel)
> Error in h(simpleError(msg, call)) :
>   error in evaluating the argument 'x' in selecting a method for function 'which': subscript out of bounds

> sessionInfo()
R version 4.0.2 (2020-06-22)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: Ubuntu 20.04 LTS

Matrix products: default
BLAS:   /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3
LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/liblapack.so.3

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=de_DE.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=de_DE.UTF-8    LC_MESSAGES=en_US.UTF-8   
 [7] LC_PAPER=de_DE.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=de_DE.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] parallel  stats     graphics  grDevices utils     datasets  methods  
[8] base     

other attached packages:
[1] FDboost_0.3-3     mboost_2.9-2      stabs_0.6-3       checkmate_2.0.0  
[5] mlr_2.17.1.9004   testthat_2.3.2    ParamHelpers_1.14

[sbrockhaus]

The error occurs in mboost:::check_newdata(newdata, blg, mf) as the number of classes of the data in newdata and mf do no match in the following check:

idx <- which(!sapply(1:length(cl1), function(i) all(cl1[[i]] == 
      cl2[[i]])))

In newdata, the functional data has class "model.matrix", "array"
In mf, the functional data has class "AsIs".
Thus, the number of class arguments do not match and we get the error in which.

[sbrockhaus]

As a quick fix, you can set all the data matrixes of functional covariates to class "AsIs":

data(fuelSubset)
fuel <- fuelSubset
fuel$UVVIS <- I(fuel$UVVIS)
fuel$NIR <- I(fuel$NIR)
fuel$dNIR <- I(t(apply(fuel$NIR, 1, diff)))
fuel$dUVVIS <- I(t(apply(fuel$UVVIS, 1, diff)))
fuel$duvvis.lambda <- fuel$uvvis.lambda[-1]
fuel$dnir.lambda <- fuel$nir.lambda[-1]

sapply(fuel, class)
str(fuel)

modH2O <- FDboost(h2o ~ bsignal(UVVIS, uvvis.lambda, knots=40, df=4)
                  + bsignal(NIR, nir.lambda, knots=40, df=4)
                  + bsignal(dUVVIS, duvvis.lambda, knots=40, df=4)
                  + bsignal(dNIR, dnir.lambda, knots=40, df=4),
                  timeformula=~bols(1), data=fuel)

# Now, the predict works
predict(modH2O, fuel)

[sbrockhaus]

Setting all functional variable matrices to class "AsIS" works now and also used to work before. So we could change the example code accordingly and add a check in FDboost() for the class of functional variables telling the user that he/she has to set functional variables to class 'AsIs'.

Or we make sure that FDboost can also handle other classes for functional variables like a numeric matrix, as it did before.

What do you think @davidruegamer @Almond-S ?

[davidruegamer]

@sbrockhaus Thanks for looking into this and your workaround. I think predict should be consistent with the original fitting behaviour, e.g., if you can input numeric matrices into FDboost to get the initial fit, you should also be able to feed them into predict function without further pre-processing.