无法从函数内的np包调用函数，可能是环境问题

Question

I am trying to use the npreg() function from the np package within another function. I am encountering an error related to environments.

npreg() is a function for non-parametric regression. I am doing the estimates in two steps, first I estimate the bandwidth using npregbw() and then I can call npreg() on the estimated bandwith to obtain the regression estimates. Outside of a function call, I encounter no issues. However, inside a function call the npreg() function does not seem to be able to work with the estimated bandwith. Reprex below:

x <- rnorm(20)
y <- 2*x + rnorm(20)
df <- data.frame(y, x)

pidtest <- function(outformula, data) {

  # estimate conditional density of outcome 
  np_lower_bw <- np::npregbw(outformula, data = data)
  np_lower <- np::npreg(np_lower_bw)
  np_lower
}

pidtest(y~x, df)

#> Error in eval(predvars, data, env): invalid 'envir' argument of type 'closure'

If I call the function just to estimate the bandwidth there is no issue

pidtest <- function(outformula, data) {

  # estimate conditional density of outcome 
  np_lower_bw <- np::npregbw(outformula, data = data)
  # np_lower <- np::npreg(np_lower_bw)
  # np_lower
  np_lower_bw
}

pidtest(y~x, df)

#> 
#> Regression Data (20 observations, 1 variable(s)):
#> 
#>                       x
#> Bandwidth(s): 0.3770171
#> 
#> Regression Type: Local-Constant
#> Bandwidth Selection Method: Least Squares Cross-Validation
#> Formula: y ~ x
#> Bandwidth Type: Fixed
#> Objective Function Value: 1.469502 (achieved on multistart 1)
#> 
#> Continuous Kernel Type: Second-Order Gaussian
#> No. Continuous Explanatory Vars.: 1

Likewise outside of a function call there is no issue:

bws <- np::npregbw(y~x, df)
np::npreg(bws)

Regression Data: 20 training points, in 1 variable(s)
                     x
Bandwidth(s): 0.307494

Kernel Regression Estimator: Local-Constant
Bandwidth Type: Fixed

Continuous Kernel Type: Second-Order Gaussian
No. Continuous Explanatory Vars.: 1

I cannot figure out why this error occurs inside my function call nor how to circumvent it. I would like to embed this estimate within a function that is doing other things so am eager to figure out a way to make it work.

Answer 1

I cannot explain exactly why, but If you try this piece of code, it works:

x <- rnorm(20)
y <- 2*x + rnorm(20)
df <- data.frame(y, x)

pidtest <- function(outformula, data) {

  # estimate conditional density of outcome 

  np_lower_bw <- np::npregbw(as.formula(outformula), data = data)
  np_lower <- np::npreg(np_lower_bw)
  np_lower
}

pidtest("y~x", df)

See here for more about the topic: https://stat.ethz.ch/pipermail/r-help/2005-March/067109.html