I want to inspect a file before processing it in R.
It is possible, that my input file is broken and contains nul values, due to a software crash. So I want my script to have a look at the data before continuing the process.
If I do a read.csv(..., skipNul = TRUE), the nuls were skipped and the script doesn't stop. But this is bad, because I don't see the lack of data.
If I do a read.csv(..., skipNul = FALSE),the nuls were skipped anyway, and I just get a warning message.
I want to count the nuls inside the file. But how can I do this?
I tried to change the encoding, but nuls are never shown when printing the dataframe.
I think, I found a solution here:
Removing nul characters
I can read the raw data by DF <- readBin("Example.log", raw(), file.info("Example.log")$size).
The resulting vector contains all 1 byte char values. I just had to filter for the nuls and print the length: length(DF[DF == 0])
Related
I have a file that contains NUL characters.
This file is generated by another program I have no control over, but I have to read it in order to get some crucial information.
Unfortunately, readChar() truncates the output with this warning:
In readChar("output.txt", 1e+05) :
truncating string with embedded nuls
Is there a way around this problem?
By convention, a text file cannot contain non-printable characters (including NUL). If a file contains such characters, it isn’t a text file — it’s a binary file.
R strictly1 adheres to this convention, and completely disallows NUL characters. You really need to read and treat the data as binary data. This means using readBin and the raw data type:
n = file.size(filename)
buffer = readBin(filename, 'raw', n = n)
# Unfortunately the above has a race condition, so check that the size hasn’t changed!
stopifnot(n == file.size(filename))
Now we can fix the buffer by removing embedded zero bytes. This assumes UTF-x or ASCII encoding! Other encodings might have embedded zero bytes that need to be interpreted!
buffer = buffer[buffer != 0L]
text = rawToChar(buffer)
1 Maybe too strictly …
This is the command I'm using :
dallData <- fread("data.csv", showProgress = TRUE, colClasses = c(rep("NULL", 2), "character", rep("NULL", 37)))
but I get this error when trying to load it: R character strings are limited to 2^31-1 bytes|
Anyway to skip those values ?
Here's a strategy that may work or at least narrow down the possible sources of error. It assumes you have enough working memory to hold the data and that your separators are really commas. If you actually have tabs as separators then you will need to modify accordingly. The plan is to read using readLines which will basically ignore the quotes that are probably mismatched. Then figure out which line or lines are at fault using count.fields, table, and which.
input <- readLines("data.csv") # ignores quotes
counts.def <- count.fields(textConnection(input),
sep=",") # defaults quotes are both ' and "
table(counts.def) # might show a variety of line counts.
# Second try with just double-quotes
counts.dbl <- count.fields(textConnection(input),
sep=",", quote="\"") # just dbl-quotes
table(counts.dbl) # if all the same, then all you do is change the quotes argument
Depending on the results you may need to edit cerain lines which can be identified using which(counts.def < 40) assuming most of them are 40 as your input efforts suggest is the expected number of fields per line.
(If the tag for [ram] means you are limited and getting warnings or using virtual memory which slows things down horribly, then you should restart your OS, and only load R before trying again. R needs contiguous block of memory and Windoze isn't very good at memory management.)
Here's a small test case to work with:
input <- readLines(textConnection(
"v1,v2,v3,v4,v5,v6
text, text, text, text, text, text
text, text, O'Malley, text,text,text
junk,junk, more junk, \"text\", tex\"t, nothing
3,4,5,6,7,8")
I've got a strange text file with a bunch of NUL characters in it (actually about 10 such files), and I'd like to programmatically replace them from within R. Here is a link to one of the files.
With the aid of this question I've finally figured out a better-than-ad-hoc way of going into each file and find-and-replacing the nuisance characters. It turns out that each pair of them should correspond to one space ([NUL][NUL]->) to maintain the intended line width of the file (which is crucial for reading these as fixed-width further down the road).
However, for robustness' sake, I prefer a more automable approach to the solution, ideally (for organization's sake) something I could add at the beginning of an R script I'm writing to clean up the files. This question looked promising but the accepted answer is insufficient - readLines throws an error whenever I try to use it on these files (unless I activate skipNul).
Is there any way to get the lines of this file into R so I could use gsub or whatever else to fix this issue without resorting to external programs?
You want to read the file as binary then you can substitute the NULs, e.g. to replace them by spaces:
r = readBin("00staff.dat", raw(), file.info("00staff.dat")$size)
r[r==as.raw(0)] = as.raw(0x20) ## replace with 0x20 = <space>
writeBin(r, "00staff.txt")
str(readLines("00staff.txt"))
# chr [1:155432] "000540952Anderson Shelley J FW1949 2000R000000000000119460007620 3 0007000704002097907KGKG1616"| __truncated__ ...
You could also substitute the NULs with a really rare character (such as "\01") and work on the string in place, e.g., let's say if you want to replace two NULs ("\00\00") with one space:
r = readBin("00staff.dat", raw(), file.info("00staff.dat")$size)
r[r==as.raw(0)] = as.raw(1)
a = gsub("\01\01", " ", rawToChar(r), fixed=TRUE)
s = strsplit(a, "\n", TRUE)[[1]]
str(s)
# chr [1:155432] "000540952Anderson Shelley J FW1949 2000R000000000000119460007620 3 0007000704002097907KGKG1616"| __truncated__
I have a large dataset in dbf file and would like to export it to the csv type file.
Thanks to SO already managed to do it smoothly.
However, when I try to import it into R (the environment I work) it combines some characters together, making some rows much longer than they should be, consequently breaking the whole database. In the end, whenever I import the exported csv file I get only half of the db.
Think the main problem is with quotes in string characters, but specifying quote="" in R didn't help (and it helps usually).
I've search for any question on how to deal with quotes when exporting in visual foxpro, but couldn't find the answer. Wanted to test this but my computer catches error stating that I don't have enough memory to complete my operation (probably due to the large db).
Any helps will be highly appreciated. I'm stuck with this problem on exporting from the dbf into R for long enough, searched everything I could and desperately looking for a simple solution on how to import large dbf to my R environment without any bugs.
(In R: Checked whether have problems with imported file and indeed most of columns have much longer nchars than there should be, while the number of rows halved. Read the db with read.csv("file.csv", quote="") -> didn't help. Reading with data.table::fread() returns error
Expected sep (',') but '0' ends field 88 on line 77980:
But according to verbose=T this function reads right number of rows (read.csv imports only about 1,5 mln rows)
Count of eol after first data row: 2811729 Subtracted 1 for last eol
and any trailing empty lines, leaving 2811728 data rows
When exporting to TYPE DELIMITED You have some control on the VFP side as to how the export formats the output file.
To change the field separator from quotes to say a pipe character you can do:
copy to myfile.csv type delimited with "|"
so that will produce something like:
|A001|,|Company 1 Ltd.|,|"Moorfields"|
You can also change the separator from a comma to another character:
copy to myfile.csv type delimited with "|" with character "#"
giving
|A001|#|Company 1 Ltd.|#|"Moorfields"|
That may help in parsing on the R side.
There are three ways to delimit a string in VFP - using the normal single and double quote characters. So to strip quotes out of character fields myfield1 and myfield2 in your DBF file you could do this in the Command Window:
close all
use myfile
copy to mybackupfile
select myfile
replace all myfield1 with chrtran(myfield1,["'],"")
replace all myfield2 with chrtran(myfield2,["'],"")
and repeat for other fields and tables.
You might have to write code to do the export, rather than simply using the COPY TO ... DELIMITED command.
SELECT thedbf
mfld_cnt = AFIELDS(mflds)
fh = FOPEN(m.filename, 1)
SCAN
FOR aa = 1 TO mfld_cnt
mcurfld = 'thedbf.' + mflds[aa, 1]
mvalue = &mcurfld
** Or you can use:
mvalue = EVAL(mcurfld)
** manipulate the contents of mvalue, possibly based on the field type
DO CASE
CASE mflds[aa, 2] = 'D'
mvalue = DTOC(mvalue)
CASE mflds[aa, 2] $ 'CM'
** Replace characters that are giving you problems in R
mvalue = STRTRAN(mvalue, ["], '')
OTHERWISE
** Etc.
ENDCASE
= FWRITE(fh, mvalue)
IF aa # mfld_cnt
= FWRITE(fh, [,])
ENDIF
ENDFOR
= FWRITE(fh, CHR(13) + CHR(10))
ENDSCAN
= FCLOSE(fh)
Note that I'm using [ ] characters to delimit strings that include commas and quotation marks. That helps readability.
*create a comma delimited file with no quotes around the character fields
copy to TYPE DELIMITED WITH "" (2 double quotes)
This is the error that I receive when I try to run tolower() on a character vector from a file that cannot be changed (at least, not manually - too large).
Error in tolower(m) : invalid multibyte string X
It seems to be French company names that are the problem with the É character. Although I have not investigated all of them (also not possible to do so manually).
It's strange, because my thought was that encoding issues would have been identified during read.csv(), rather than during operations after the fact.
Is there a quick way to remove these multibyte strings? Or, perhaps a way to identify and convert? Or even just ignore them entirely?
Here's how I solved my problem:
First, I opened the raw data in a texteditor (Geany, in this case), clicked properties and identified the Encoding type.
After which I used the iconv() function.
x <- iconv(x,"WINDOWS-1252","UTF-8")
To be more specific, I did this for every column of the data.frame from the imported CSV. Important to note that I set stringsAsFactors=FALSE in my read.csv() call.
dat[,sapply(dat,is.character)] <- sapply(
dat[,sapply(dat,is.character)],
iconv,"WINDOWS-1252","UTF-8")
I was getting the same error. However, in my case it wasn't when I was reading the file, but a bit later when processing it. I realised that I was getting the error, because the file wasn't read with the correct encoding in the first place.
I found a much simpler solution (at least for my case) and wanted to share. I simply added encoding as below and it worked.
read.csv(<path>, encoding = "UTF-8")
library(tidyverse)
data_clean = data %>%
mutate(new_lowercase_col = tolower(enc2utf8(as.character(my_old_column))))
Where new_lowercase_col is the name of the new column I'm making out of the old uppercase one, which was called my_old_column.
I know this has been answered already but thought I'd share my solution to this as I experienced the same thing.
In my case, I used the function str_trim() from package stringr to trim whitespace from start and end of string.
com$uppervar<-toupper(str_trim(com$var))
# to avoid datatables warning: error in tolower(x) invalid multibyte string
# assuming all columns are char
new_data <- as.data.frame(
lapply(old_data, enc2utf8),
stringsAsFactors = FALSE
)
My solution to this issue
library(dplyr) # pipes
library(stringi) # for stri_enc_isutf8
#Read in csv data
old_data<- read.csv("non_utf_data.csv", encoding = "UTF-8")
#despite specifying utf -8, the below columns are not utf8:
all(stri_enc_isutf8(old_data$problem_column))
#The below code uses regular expressions to cleanse. May need to tinker with the last
#portion that selects the grammar to retain
utf_eight_data<- old_data %>%
mutate(problem_column = gsub("[^[:alnum:][:blank:]?&/\\-]", "", old_data$problem_column)) %>%
rename(solved_problem = problem_column)
#this column is now utf 8.
all(stri_enc_isutf8(utf_eight_data$solved_problem))