Question

I’m trying to write a predict style function for lme4’s lmer models. The idea is that the predict function will work on a dataframe containing factor and variable levels, of the kind produced by expand.grid:

level_df <- expand.grid(a=1:3, b=4:6)

I’m at the point where I can get the formula for calculating a given effect/interaction as a string:

formula_string <- "a * b + 3"

But the only way I’ve been able to apply that formula to the level dataframe (in this case, evaluating the formula for each value of a and b) is through some pretty liberal and probably dangerous use of things like assign and eval:

EffectFunction <- function(df_row, formula_string) {
  l_cols <- colnames(df_row)
  for (i in 1:ncol(df_row)) {
    assign(l_cols[i], df_row[1, i])
  }
  parsed <- parse(text=formula_string)
  df_row$effect_val <- eval(parsed)
  return(df_row)
}

pred_results <- adply(
  level_df,
  1,
  formula_string=formula_string,
  EffectFunction
)

Are there safer and faster ways to convert between strings and symbols/variable names?

Answer 1

You can use a data.frame as the envir argument for eval

so

 eval(parse(text = formula_string), envir = level_df)

should work nicely as it will look (first) within level_df for any variables.