Collection of non-meaningful error messages / handling

[davidruegamer]

• For SOF models, which are specified with a data list containing a response with a wrong length (e.g. response has length 7 and covariates have length 8 or 8 rows, FDboost does not throw an error until the df2lambda function tries to do the cholesky decomposition with wrong weights (credits to Joanna)

fuelSubset$heatan <- fuelSubset$heatan[1:100]
sof <- FDboost(heatan ~ bsignal(NIR, nir.lambda, knots = 20, df = 4),
                         timeformula = NULL, data = fuelSubset)

[davidruegamer]

Not saying, that we should deal with all of those errors, but some may improve the package handling:

• Error handling when defining knots = 0 : The first line does meaningful things, the second doesn't:

str(with(fuelSubset, bsignal(NIR, nir.lambda, knots = 0)),1)

sof <- FDboost(heatan ~ bbs(h2o, df = 4)
                 + bsignal(UVVIS, uvvis.lambda, knots = 20, df = 4)
                 + bsignal(NIR, nir.lambda, knots = 0, df = 4),
                     timeformula = NULL, data = fuelSubset)

Why is that? Our problem? EDIT: It is possible to construct a design matrix with zero knots but it is not possible to invert t(X)%*%X in this case. So I think this isn't our problem...

• Better error handling for missing timeformula ? Set a default?
• When forgetting the tilde in timeformula, complete.cases throws a quite mysterious error message:

fos_intercept <- FDboost(EMG ~ 1,
                             timeformula = bbs(t, df = 2),
                             data = emotion)

• Same for id in the long format case. FDboost reports, that it can not find the id object, instead of reporting a wrong specified argument:

fos_intercept_long <- FDboost(EMG ~ 1,
                                 timeformula = ~ bbs(time_long, df = 4),
                                 id = curveid, data = emotion_long)

[davidruegamer]

• Dimension problem with non-meaningful error message also present for FOF regression