[R] Grep with wildcards across multiple columns

[Bush, Daniel P. DPI]

I think the way I set up my sample data without any explanation confused things slightly. These data might make things clearer:

# Create fake data
df <- data.frame(code   = c(rep(1001, 8), rep(1002, 8)),
                 year   = rep(c(rep(2011, 4), rep(2012, 4)), 2),
                 fund   = rep(c("10E", "27E", "27E", "29E"), 4),
                 func   = rep(c("110000", "122000", "214000", "158000"), 4),
                 obj    = rep(c("100", "100", "210", "220"), 4),
                 amount = round(rnorm(16, 50000, 10000)))

These are financial data with a hierarchical account structure where a zero represents a summary account that rolls up all the accounts at subsequent digits (e.g. 100000 rolls up 110000, 122000, 158000, etc.). I was trying to do two things with the search parameters: turn zeroes into question marks, and duplicate the functionality of a SQL query using those question marks as wildcards:

# Set parameters
par.fund <- "20E"; par.func <- "100000"; par.obj <- "000"
par.fund <- glob2rx(gsub("0", "?", par.fund))
par.func <- glob2rx(gsub("0", "?", par.func))
par.obj <- glob2rx(gsub("0", "?", par.obj))

Fortunately, Bill's suggestion to use the intersect function worked just fine--since intersect accepts only two arguments, I had to nest a pair of statements:

# Solution: Use a pair of nested intersects
dt2 <- dt[intersect(intersect(grep(par.fund, fund), grep(par.func, func)),
                    grep(par.obj, obj)),
          sum(amount), by=c('code', 'year')]
df2 <- ddply(df[intersect(intersect(grep(par.fund, df$fund),
                                    grep(par.func, df$func)),
                          grep(par.obj, df$obj)), ],
             .(code, year), summarize, amount = sum(amount))

Thanks for your ideas!

DB

Daniel Bush | School Finance Consultant 
School Financial Services | Wis. Dept. of Public Instruction 
daniel.bush -at- dpi.wi.gov | 608-267-9212

-----Original Message-----
From: William Dunlap [mailto:wdunlap at tibco.com] 
Sent: Thursday, March 14, 2013 5:49 PM
To: Bush, Daniel P. DPI; 'r-help at r-project.org'
Subject: RE: Grep with wildcards across multiple columns

grep(pattern, textVector) returns of the integer indices of the elements of textVector that match the pattern.  E.g.,
  > grep("T", c("One","Two","Three","Four"))
  [1] 2 3

The '&' operator naturally operates on logical vectors of the same length (If you give it numbers it silently converts 0 to FALSE and  other numbers to TRUE.)

The two don't fit together.  You could use grepl(), which returns a logical vector the length of textVector, as in
   grepl(p1,v1) & grepl(p2,v2)
to figure which entries in the table have v1 matching p1 and v2 matching p2.

Or, you could use
  intersect(grep(p1,v1), grep(p2,v2))
if you want to stick with integer indices.

Bill Dunlap
Spotfire, TIBCO Software
wdunlap tibco.com


> -----Original Message-----
> From: r-help-bounces at r-project.org 
> [mailto:r-help-bounces at r-project.org] On Behalf Of Bush, Daniel P. DPI
> Sent: Thursday, March 14, 2013 2:43 PM
> To: 'r-help at r-project.org'
> Subject: [R] Grep with wildcards across multiple columns
> 
> I have a fairly large data set with six variables set up like the following dummy:
> 
> # Create fake data
> df <- data.frame(code   = c(rep(1001, 8), rep(1002, 8)),
>                  year   = rep(c(rep(2011, 4), rep(2012, 4)), 2),
>                  fund   = rep(c("10E", "10E", "10E", "27E"), 4),
>                  func   = rep(c("110000", "122000", "214000", "158000"), 4),
>                  obj    = rep("100", 16),
>                  amount = round(rnorm(16, 50000, 10000)))
> 
> What I would like to do is sum the amount variable by code and year, 
> filtering rows using different wildcard searches in each of three 
> columns: "1?E" in fund, "1??????" in func, and "???" in obj. I'm OK turning these into regular expressions:
> 
> # Set parameters
> par.fund <- "10E"; par.func <- "100000"; par.obj <- "000"
> par.fund <- glob2rx(gsub("0", "?", par.fund)) par.func <- 
> glob2rx(gsub("0", "?", par.func)) par.obj <- glob2rx(gsub("0", "?", 
> par.obj))
> 
> The problem occurs when I try to apply multiple greps across columns. 
> I'd prefer to use data.table since it's so much faster than plyr and I 
> have 159 different sets of parameters to run through, but I get the same error setting it up either way:
> 
> # Doesn't work
> library(data.table)
> dt <- data.table(df)
> eval(parse(text=paste(
>   "dt2 <- dt[", "grep('", par.fund, "', fund) & ",
>   "grep('", par.func, "', func) & grep('", par.obj, "', obj)",
>   ", sum(amount), by=c('code', 'year')]" , sep=""))) # Warning 
> message:
> #   In grep("^1.E$", fund) & grep("^1.....$", func) :
> #   longer object length is not a multiple of shorter object length
> 
> # Also doesn't work
> library(plyr)
> eval(parse(text=paste(
>   "df2 <- ddply(df[grep('", par.fund, "', df$fund) & ",
>   "grep('", par.func, "', df$func) & grep('", par.obj, "', df$obj), ]",
>   ", .(code, year), summarize, amount = sum(amount))" , sep=""))) # 
> Warning message:
> #   In grep("^1.E$", df$fund) & grep("^1.....$", df$func) :
> #   longer object length is not a multiple of shorter object length
> 
> Clearly, the problem is how I'm trying to combine greps in subsetting 
> rows, but I haven't been able to find a solution that works. Any 
> thoughts-preferably something that works with data.table?
> 
> DB
> 
> Daniel Bush
> School Finance Consultant
> School Financial Services
> Wisconsin Department of Public Instruction PO Box 7841 | Madison, WI 
> 53707-7841 daniel.bush -at- dpi.wi.gov | sfs.dpi.wi.gov
> Ph: 608-267-9212 | Fax: 608-266-2840