Is there a way to add methods to a function within an R package?

Question

I am trying to use lime to add ML explanations in the output of a package I am developing. My solution uses a gradient boosting model from library gbm. This type of model is not supported by lime as is, so I would need to add a gbm method to generics: -model_type -predict_model

if I do this manually, running my code line by line, I have no problem.
However, once lime is included in a function and I try to run it, it looks like the function cannot assign the "gbm" method. .

Would anybody have an idea about how to approach this?

I have also tried using setMethod, but then I would get a message saying the environment is locked

I train my model outside the package function:

    lime_model<-gbm(train_target ~. ,
                 data = train,
                 distribution="bernoulli",
                 n.trees = N_trees,
                 interaction.depth = Int.depth,
                 shrinkage=Learn_rate,
                 n.minobsinnode=Min.obs.node)

then pass it to the function:

    explanation<-        
    lime_gbm(train_data=train,test_data=test,model=lime_model,Conf, 
                            n_trees=N_trees)

defined as:

    #' @param train_data a dataframe containing the data to train the model on
    #' @param test_data a dataframe containing the data to test test the model on
    #' @param Conf a config file to extract general variable names
    #' @param n_trees number of trees in the model
    #' @param model the model to be explained
    #' 
    #' @import dplyr
    #' @import lime 
    #'
    #' @return list
    #' @export

    lime_gbm<-function(train_data,test_data,model,Conf, n_trees){

      if(!is.data.frame(train_data) | !is.data.frame(test_data)){
        stop("data must be a data.frame")
      }

      if(!is.numeric(n_trees)){
        stop("model parameters must be numeric")
      }

      if(!class(model)=="gbm"){
        stop("model must be of class gbm")
      }

      fit.gbm=model

      #create model type for gbm
      model_type.gbm <- function(x,...){
        return("classification")
      }

      #setMethod("model_type","gbm",function(x,...) "classification")

      predict_model.gbm <- function(x, newdata, type="prob",...) {

        res <- predict.gbm(x,newdata = newdata, type="response",
                       n.trees=n_trees)
        res <-as.data.frame(res)

        colnames(res)<-"predictions"#c("Yes","No")
        return(res)

      }
      #setMethod("predict_model","gbm",function(x,...) as.data.frame(x))

      explainer <- lime::lime(train_data, 
                              fit.gbm, 
                              bin_continous=T, 
                              quantile_bins=F)

     explanation <- lime::explain(test_data,
                                   explainer,
                                   n.trees=n_trees,
                                   n_labels=1,
                                   #model_type="classification",
                                   n_features=ncol(test_data),
                                   n_permutations = 10,#500, 
                                   feature_select='lasso_path',
                                   dist_fun="manhattan")

      return(explanation)
    }

The code as is gives the error message: "Error: The class of model must have a model_type method. See ? model_type to get an overview of models supported out of the box"

or, if I remove ".gbm" when assigning the method and remove # in front of

   #setMethod()

:

Error in setMethod("model_type", "gbm", function(x, ...) "classification") : the environment ‘SiteViewModelling’ is locked; cannot assign methods for function ‘model_type’

Is there a way to add methods to a function within an R package?

0 Answers0