greenwoodlab / pcev Goto Github PK

At the moment, interaction terms need to be included explicitly in the PCEV framework, i.e. covariate needs to include a column corresponding to the multiplication of other variables. Can we do better? I think that would require a reformatting of the front-end, perhaps including a formula-type interface...

I've recently received questions about the output of computePCEV and I realized that I never explain the different components of an object of class Pcev. It would probably make the most sense to add this directly to the documentation page for computePCEV, since this is the main function of the package.

At the moment, setting nperm = 0 works, and it behaves as expected (i.e. the model is fitted and no permutation is performed. However, it triggers an error in the print method. We need to replace the p-value with NA or something.

In my most recent simulations (and also in the SMMR paper), I choose the blocks randomly, which still gives a decent performance. So I thought I could add that option in the package, in two ways:

1. If index is a single positive integer, this is assumed to be the number of blocks.

2. If index = "adaptive", then the number of blocks is chosen to be 2*p/n, so that there are about n/2 variables per block.

In boths cases, the variables are assigned to a given block at random.

Hello. When using the pcev package, I get an error when the number of covariates (q) is larger than the number of response variables (p). As you can see in the output below, the error message I get when calling computePCEV(Y, X) is "Error in weights[, 2:(ncol(pcevObj$X) - 1), drop = FALSE]: subscript out of bounds". Is there a reason why PCEV shouldn't be used when q > p, or is there an error in the package? Thank you.

# Attach pcev package
    library(pcev)
    
    # Define dimensions
    p <- 2 # Number of response variables
    q <- 3 # Number of covariates
    n <- 100 # Number of observations
    
    # Generate response matrix and covariate matrix
    set.seed(10)
    Y <- matrix(rnorm(n * p), nrow = n, ncol = p) # Response variables
    X <- matrix(rnorm(n * q), nrow = n, ncol = q) # Covariates
    
    # Attempt to perform PCEV on Y and X
    pcev_out <- computePCEV(Y, X)
#> Error in weights[, 2:(ncol(pcevObj$X) - 1), drop = FALSE]: subscript out of bounds

^{Created on 2020-07-06 by the reprex package (v0.3.0)}