Support rstanarm in add_predictions

I expect when calling add_predictions on a dataframe with an rstanarm model, that it will add a single new column, with default name pred, to the input dataframe containing the predictions of applying the model. Instead it raises an error.

Running:

model <- rstanarm::stan_glm(x ~ 1 , data=data.frame(x=c(1,2)), refresh=0)
modelr::add_predictions(data.frame(x=1), model)

Result:

Error in `[[<-.data.frame`(`*tmp*`, var, value = c(`1` = 1.5023766891339, : replacement has 2 rows, data has 1
Traceback:

1. modelr::add_predictions(data.frame(x = 1), model)
2. `[[<-`(`*tmp*`, var, value = c(`1` = 1.5023766891339, `2` = 1.5023766891339
 . ))
3. `[[<-.data.frame`(`*tmp*`, var, value = c(`1` = 1.5023766891339, 
 . `2` = 1.5023766891339))
4. stop(sprintf(ngettext(N, "replacement has %d row, data has %d", 
 .     "replacement has %d rows, data has %d"), N, nrows), domain = NA)

The reason is predict.stanreg ignores all positional arguments after the first, and requires the newdata argument to be named. However add_predictions passes the data by position.

A workaround would be to change the predict2 function in predictions.R to explicitly pass the data as argument newdata. But I'm not sure if this causes issues with other models.

predict2 <- function(model, data, type = NULL) {
  if (is.null(type)) {
    stats::predict(model, newdata = data)
  } else {
    stats::predict(model, newdata = data, type = type)
  }
}

tidyverse / modelr

Support rstanarm in add_predictions #113