explain() fails uninformatively if exact argument is specified, on cv.glmnet model objects

[nimirea]

When trying to use fastshap with a cv.glmnet model, specifying exact argument to fastshap::explain() causes the function to fail, no matter what the supplied value of exact is. The error returned is not very helpful.

Here's the documentation on exact (emphasis mine):

Logical indicating whether to compute exact Shapley values. Currently only available for stats::lm(), xgboost::xgboost(), and lightgbm::lightgbm() objects. Default is FALSE. Note that setting exact = TRUE will return explanations for each of the stats::terms() in an stats::lm() object. Default is FALSE.

Based on this, I would expect the following behavior when calling fastshap::explain() with an explicit exact argument:

• fastshap::explain(..., exact = FALSE) should produce the default behavior (i.e., same as calling without explicitly setting exact), since the documentation states the default value for exact is FALSE.
• fastshap::explain(..., exact = TRUE) should:
  • produce exact Shapley values for stats::lm(), xgboost::xgboost(), and lightgbm::lightgbm() objects
  • fail with an informative error otherwise (e.g., "You have set exact = TRUE for an object type that does not support this functionality, please see ?fastshap::explain() for more details")

Instead, both cases yielded this error:

Error in `colnames<-`(`*tmp*`, value = feature_names) :
  attempt to set 'colnames' on an object with less than two dimensions

The function only works when exact is left unspecified.

Here's a reprex (kudos to @DavZim for the example this is based on):

library(fastshap)
library(AmesHousing)
library(glmnet)

data <- AmesHousing::make_ames()

y <- data$Sale_Price
X <- as.matrix(data[, c("Lot_Frontage", "Lot_Area", "Year_Sold", "Year_Built")])

glm <- cv.glmnet(X, y)
pred_wrapper_fun <- function(mdl, newdata) as.numeric(predict(mdl, newdata, s = glm$lambda.min))

# This works:
expl_glm <- fastshap::explain(
  object = glm,
  X = X,
  pred_wrapper = pred_wrapper_fun
)

# This doesn't work:
expl_glm <- fastshap::explain(
  object = glm,
  X = X,
  pred_wrapper = pred_wrapper_fun,
  exact = FALSE
)
# Error in `colnames<-`(`*tmp*`, value = feature_names) :
#  attempt to set 'colnames' on an object with less than two dimensions

# This also doesn't work:
expl_glm <- fastshap::explain(
  object = glm,
  X = X,
  pred_wrapper = pred_wrapper_fun,
  exact = TRUE
)
# Error in `colnames<-`(`*tmp*`, value = feature_names) :
#  attempt to set 'colnames' on an object with less than two dimensions

# This also doesn't work:
expl_glm <- fastshap::explain(
  object = glm,
  X = X,
  pred_wrapper = pred_wrapper_fun,
  exact = NULL
)
# Error in `colnames<-`(`*tmp*`, value = feature_names) :
#  attempt to set 'colnames' on an object with less than two dimensions

I am not sure whether this happens with other types of models for which exact = TRUE is not supported.

[bgreenwell]

Hi @nimirea, thanks for the detailed example. I should be able to look at the source code and push a fix shortly after the holiday!

In the meantime, which version of fastshap are you using?

[nimirea]

Thanks! Version is 4.3.0.

[bgreenwell]

Hi @nimirea, is that the R version? Let me know which fastshap version you have installed when you can.