The extent of the data rectangle can be determined in various ways:
read_excel()uses the smallest rectangle that contains the non-empty cells. It “shrink wraps” the data.
read_excel()’s discovery process with respect to rows. At least
skipspreadsheet rows will be skipped or ignored and at most
n_maxspreadsheet rows will be considered as data. Compared to the default of discovery, these arguments can only lead to making the output tibble smaller.
rangeargument is taken literally, even if that means you will have leading or trailing rows or columns filled with
NA. If you ask for
range = "A1:D4", you are guaranteed to get a tibble with 4 columns (A through D) and either 3 rows (
col_names = TRUE, default) or 4 rows (
col_names = FALSE).
read_excel()’s geometry arguments often imply that certain limits are discovered while others are bounded or set. This will be more clear in the concrete examples below.
For now, here are a few ways
read_excel() can look when you take control of the geometry:
read_excel("yo.xlsx", skip = 5) read_excel("yo.xlsx", n_max = 100) read_excel("yo.xlsx", skip = 5, n_max = 100) read_excel("yo.xlsx", range = "C1:E7") read_excel("yo.xlsx", range = cell_rows(6:23)) read_excel("yo.xlsx", range = cell_cols("B:D")) read_excel("yo.xlsx", range = anchored("C4", dim = c(3, 2)))
readxl’s behavior and interface may be easier to understand if you understand this about Excel:
Cells you can see don’t necessarily exist. Cells that look blank aren’t necessarily so.
Among lots of other information, Excel files obviously must contain information on each cell. Let’s use the word “item” to denote one cell’s-worth of info.
Just because you see a cell on the screen in Excel, that doesn’t mean there’s a corresponding item on file. Why? Because Excel presents a huge gridded canvas for you to write on. Until you actually populate a cell, though, it doesn’t really exist.
The stream of cell items describes the existing cells, going from upper left to lower right, travelling by row. Blank cells simply do not exist in it.
Ah, but what is a blank cell? Some cells appear blank to the naked eye but are not considered so by Excel and, indeed, are represented by a cell item. This happens when a cell has no content but does have an associated format. This format could have been applied directly to a single cell or, more often, indirectly via formatting applied to an entire row or column. Once a human has spent some quality time with a spreadsheet, many seemingly empty cells will bear a format and will thus have an associated cell item.
readxl only reads cell items that have content. It ignores cell items that exist strictly to convey formatting.
The tibble returned by readxl will often cover cells that are empty in the spreadsheet, filled with
NA. But only because there was some other reason for the associated row or column to exist: actual data or user-specified geometry.
n_max are the “entry-level” solution for controlling the data rectangle. They work only in the row direction. Column-wise, you’re letting readxl discover which columns are populated.
If you specify
range (covered below),
n_max are ignored.
skip argument tells
read_excel() to start looking for populated cells after skipping at least
skip rows. If the new start point begins with 1 or more empty rows,
read_excel() will skip even more before it starts reading from the sheet.
Here’s a screen shot of the
geometry.xlsx example sheet that ships with readxl, accessible via
read_excel() just discovers the data rectangle:
read_excel(readxl_example("geometry.xlsx")) #> # A tibble: 3 x 3 #> B3 C3 D3 #> <chr> <chr> <chr> #> 1 B4 C4 D4 #> 2 B5 C5 D5 #> 3 B6 C6 D6
If you explicitly skip one row, note that
read_excel() still skips row 2, which is also empty, leading to the same result as before:
read_excel(readxl_example("geometry.xlsx"), skip = 1) #> # A tibble: 3 x 3 #> B3 C3 D3 #> <chr> <chr> <chr> #> 1 B4 C4 D4 #> 2 B5 C5 D5 #> 3 B6 C6 D6
You can also use
skip to skip over populated cells. In real life, this is a mighty weapon against the explanatory text that people like to include at the top of spreadsheets.
read_excel(readxl_example("geometry.xlsx"), skip = 3) #> # A tibble: 2 x 3 #> B4 C4 D4 #> <chr> <chr> <chr> #> 1 B5 C5 D5 #> 2 B6 C6 D6
read_excel() to skip at least this many spreadsheet rows before reading anything.
n_max argument tells
read_excel() to read at most
n_max rows, once it has found the data rectangle. Note that
n_max is specifically about the data. You still use
col_names to express whether the first spreadsheet row should be used to create column names (default is
n_max = 2 causes us to ignore the last data row – the 3rd one – in
read_excel(readxl_example("geometry.xlsx"), n_max = 2) #> # A tibble: 2 x 3 #> B3 C3 D3 #> <chr> <chr> <chr> #> 1 B4 C4 D4 #> 2 B5 C5 D5
n_max is an upper bound. It will never cause empty rows to be included in the tibble. Note how we get 3 data rows here, even though
n_max is much greater.
range argument is the most flexible way to control geometry and is powered by the cellranger package.
One huge difference from
n_max is that
range is taken literally! Even if it means the returned tibble will have entire rows or columns consisting of
You can describe cell limits in a variety of ways:
Excel-style range: Specify a fixed rectangle with
range = "A1:D4" or
range = "R1C1:R4C4". You can even prepend the worksheet name like so:
range = "foofy!A1:D4" and it will be passed along to the
deaths.xlsx example sheet features junk rows both before and after the data rectangle. The payoff for specifying the data rectangle precisely is that we get the data frame we want, with correct guesses for the column types.
read_excel(readxl_example("deaths.xlsx"), range = "arts!A5:F15") #> # A tibble: 10 x 6 #> Name Profession Age `Has kids` `Date of birth` `Date of death` #> <chr> <chr> <dbl> <lgl> <dttm> <dttm> #> 1 David Bow… musician 69 TRUE 1947-01-08 00:00:00 2016-01-10 00:00:00 #> 2 Carrie Fi… actor 60 TRUE 1956-10-21 00:00:00 2016-12-27 00:00:00 #> 3 Chuck Ber… musician 90 TRUE 1926-10-18 00:00:00 2017-03-18 00:00:00 #> 4 Bill Paxt… actor 61 TRUE 1955-05-17 00:00:00 2017-02-25 00:00:00 #> # … with 6 more rows
We repeat the screenshot of
geometry.xlsx as a visual reference.
Going back to
geometry.xlsx, here we specify a rectangle that only partially overlaps the data. Note the use of default column names, because the first row of cells is empty, and the leading column of
read_excel(readxl_example("geometry.xlsx"), range = "A2:C4") #> New names: #> * `` -> ...1 #> * `` -> ...2 #> * `` -> ...3 #> # A tibble: 2 x 3 #> ...1 ...2 ...3 #> <lgl> <chr> <chr> #> 1 NA B3 C3 #> 2 NA B4 C4
Specific range of rows or columns: Set exact limits on just the rows or just the columns and allow the limits in the other direction to be discovered. Example calls:
## rows only read_excel(..., range = cell_rows(1:10)) ## is equivalent to read_excel(..., range = cell_rows(c(1, 10))) ## columns only read_excel(..., range = cell_cols(1:26)) ## is equivalent to all of these read_excel(..., range = cell_cols(c(1, 26))) read_excel(..., range = cell_cols("A:Z")) read_excel(..., range = cell_cols(LETTERS)) read_excel(..., range = cell_cols(c("A", "Z"))
geometry.xlsx to demonstrate setting hard limits on the rows, running past the data, while allowing column limits to discovered. Note the trailing rows of
read_excel(readxl_example("geometry.xlsx"), range = cell_rows(4:8)) #> # A tibble: 4 x 3 #> B4 C4 D4 #> <chr> <chr> <chr> #> 1 B5 C5 D5 #> 2 B6 C6 D6 #> 3 <NA> <NA> <NA> #> 4 <NA> <NA> <NA>
Here we get a 3 by 4 rectangle with cell C5 as the upper left corner:
read_excel( readxl_example("geometry.xlsx"), col_names = paste("var", 1:4, sep = "_"), range = anchored("C5", c(3, 4)) ) #> # A tibble: 3 x 4 #> var_1 var_2 var_3 var_4 #> <chr> <chr> <lgl> <lgl> #> 1 C5 D5 NA NA #> 2 C6 D6 NA NA #> 3 <NA> <NA> NA NA
Here we set C5 as the upper left corner and allow the other limits to be discovered: