Debugging Example

• Video lecture: https://youtu.be/-P9UzQuJSH8
• This document and all example code: http://winvector.github.io/Debugging/
• Blog announcement (good place for questions and comments): http://www.win-vector.com/blog/2016/04/free-data-science-video-lecture-debugging-in-r/

## Warning: package 'ggplot2' was built under R version 3.2.4

## Loading required package: gplots

## Warning: package 'gplots' was built under R version 3.2.4

## 
## Attaching package: 'gplots'

## The following object is masked from 'package:stats':
## 
##     lowess

# debug runner from https://github.com/WinVector/WVPlots/blob/master/R/DebugFn.R
#' Capture arguments of exception throwing plot for later debugging.
#'
#' Run fn and print result, save arguments on failure.  Use on systems like ggplot()
#' where some calculation is delayed until print().
#'
#' @param saveFile path to save RDS to.
#' @param fn function to call
#' @param ... arguments for fn
#' @return fn(...) normally, but if f(...) throws an exception save to saveFile RDS of list r such that do.call(r$fn,r$args) repeats the call to fn with args.
#'
#' @examples
#'
#' d <- data.frame(x=1:5)
#' DebugPrintFn('problem.RDS','PlotDistCountNormal',d,xvar='x','example')
#' tryCatch(
#'    DebugPrintFn('problem.RDS','PlotDistCountNormal',
#'       d,xvar='xmisspelled','example'),
#'    error = function(e) { print(e) })
#'
#' @export
DebugPrintFn <- function(saveFile,fn,...) {
  args <- list(...)
  tryCatch({
    res = do.call(fn,args)
    print(res)
    res
  },
  error = function(e) {
    saveRDS(object=list(fn=fn,args=args),file=saveFile)
    stop(paste0("Wrote '",saveFile,"' on catching '",as.character(e),"'"))
  })
}

# Run once, seems okay
set.seed(25352)
d <- pretendBigExpensiveExperiment()
print(ROCPlot(d,'x','y','example'))

# Run a lot of important experiment
for(runNum in seq_len(100)) {
  d <- pretendBigExpensiveExperiment()
  print(ROCPlot(d,'x','y',paste('example',runNum)))
}

## Error in ROCR::prediction(predcol, outcol): Number of classes is not equal to 2.
## ROCR currently supports only evaluation of binary classification tasks.

# Run a lot of important experiment
# not the exact same set of experiments as we didn't reset pseudo-random seed!
for(runNum in seq_len(100)) {
  d <- pretendBigExpensiveExperiment()
  DebugPrintFn('problem.RDS','ROCPlot',d,'x','y',paste('example',runNum))
}

## Error in value[[3L]](cond): Wrote 'problem.RDS' on catching 'Error in ROCR::prediction(predcol, outcol): Number of classes is not equal to 2.
## ROCR currently supports only evaluation of binary classification tasks.
## '

problem <- readRDS('problem.RDS')
print(problem)

## $fn
## [1] "ROCPlot"
## 
## $args
## $args[[1]]
##             x     y
## 1  1.14402934 FALSE
## 2 -0.68807307 FALSE
## 3  0.05036758 FALSE
## 4 -0.72399384 FALSE
## 5  0.72194744 FALSE
## 
## $args[[2]]
## [1] "x"
## 
## $args[[3]]
## [1] "y"
## 
## $args[[4]]
## [1] "example 34"

do.call(problem$fn,problem$args)

## Error in ROCR::prediction(predcol, outcol): Number of classes is not equal to 2.
## ROCR currently supports only evaluation of binary classification tasks.