Error in predict()

[Former user]

Hi,

I have an spls-regression model and use predict() on new X data. This gernerally works fine, but fails when X has only one row: "Error ... attempt to set 'rownames' on an object with no dimensions"

Debugging shows that the error occurs in the line

rownames(Y.hat[[i]]) = rownames(newdata[[i]])

rownames(Y.hat[[i]])

throws a "NULL" in the 1 row case.

Any idea what is going on here?

[Florian Rohart]

Hi there,

When you say "X has only one row", is that the 'newdata' or the original X of the spls model?

I can't replicate the problem when 'newdata has only one row (and is a matrix as required)

[Sally Weiss]

Hi,

sorry for the confusion.

It fails here , if 'newdata' is a one row matrix. I tried to use different one-row 'newdata', but now difference.

However, I observed this in version 6.1.0, but not in 5.2.0.

[Florian Rohart]

Hi Sally,

Could you provide me with a minimal example that reproduces the problem? I've tried on mixOmics v6.1.0 (the 6.1.1 is out since last October, few bugs fixed) and everything is working fine - see example below with some data available in mixOmics. If it's because of some specificities of your data, you can send us part of your data by email that led to the problem and it will only be used for debugging purposes -everything will be confidential.

data(liver.toxicity)
X <- liver.toxicity$gene
Y <- liver.toxicity$clinic

toxicity.spls <- spls(X, Y, ncomp = 2, keepX = c(50, 50), keepY = c(10, 10))

newdata = matrix(rnorm(3116), nrow=1) # newdata must be a matrix of one row and 3116 columns as X
colnames(newdata) = colnames(X) # 'newdata' must include all the variables of 'object$X'

pred = predict(toxicity.spls, newdata)