In matchmaker: Flexible Dictionary-Based Cleaning

Working with Data Frames

When using the match_df() function, you would construct the dictionary same as you would above, with two extra columns that specify the column name in the data frame and the order the resulting values should be (if the column is a factor).

As with match_vec(), all the same keywords apply, but now there are also two keywords for the columns:

• .regex [pattern]: any column whose name is matched by [pattern]. The [pattern] should be an unquoted, valid, PERL-flavored regular expression. This will match any column that is named with a given pattern. This would commonly be used for recoding results from columns that all start with the same pattern: ^lab_result_ would match lab_result_QTPCR, lab_result_WBC, lab_result_iron.
• .global: defines rules for any column that is a character or factor and any column named in the dictionary. If you want to apply a set of definitions to all valid columns in addition to specified columns, then you can include a .global group in the by column of your ‘dictionary’ data frame. This is useful for setting up a dictionary of common spelling errors. NOTE: specific variable definitions will override global defintions. For example: if you have a column for cardinal directions and a definiton for N = North, then the global variable N = no will not override that.

Matching columns with .regex

Before you use regex, you should be aware of three special symbols that will help anchor your words and prevent any unintended matching.

1. The carrot (^) should be placed at the beginning of a pattern to show that it's the beginning of the word. For example, lab will match both lab_result and granite_slab, but ^lab will only match lab_result
2. The dollar ($) should be placed at the end of a pattern to show that it's the end of a word. For example, date will match both admission_date and date_of_onset, but date$ will only match admission_date$.
3. The dot (.) matches any character. Because it's common in column names imported by R, it's a good idea to wrap it in square brackets ([.]) to tell R that you actually mean a dot. For example, ^lab.r$ will match lab.r, lab_r, and labor, but ^lab[.]r$ will only match lab.r.

The best strategy is to use at least one anchor to prevent it greedily selecting columns to match.

In our example from the top, there are three columns that all start with lab_result_, so we use the .regex ^lab_result keyword:

# view the lab_result columns:
print(labs <- grep("^lab_result_", names(dat), value = TRUE))
str(dat[labs])
# show the lab_result part of the dictionary:
print(dict[grep("^[.]regex", dict$grp), ])
# clean the data and compare the result
cleaned <- match_df(dat, dict, 
  from = "options", 
  to = "values", 
  by = "grp", 
  order = "orders"
) 
str(cleaned[labs])

Using .global to clean up all character/factor columns

We've actually seen the .global keyword in use already. Let's take one more look at the results from above:

# show the lab_result part of the dictionary:
print(dict[grep("^[.]regex", dict$grp), ])
# show the original data
str(dat[labs])
# show the modified data
str(cleaned[labs])

Notice above how there are rules for "high", "norm", and "inc", but not for "unk", which was turned into "unknown"? This is because of the global keywords:

print(dict[grep("^[.](regex|global)", dict$grp), ])

The "unk" keyword was defined in our global dictionary and has been used to translate "unk" to "unknown".

Of course, be very careful with this one.

Warnings

Internally, the match_vec() function can be quite noisy with warnings for various reasons. Thus, by default, the match_df() function will keep these quiet, but you can have them printed to your console if you use the warn = TRUE option:

cleaned <- match_df(dat, dict, 
  from = "options", 
  to = "values", 
  by = "grp", 
  order = "orders",
  warn = TRUE
) 

matchmaker documentation built on Feb. 22, 2020, 1:11 a.m.