I have the following "stacked JSON" object within R, example1.json:
{"ID":"12345","Timestamp":"20140101", "Usefulness":"Yes",
"Code":[{"event1":"A","result":"1"},…]}
{"ID":"1A35B","Timestamp":"20140102", "Usefulness":"No",
"Code":[{"event1":"B","result":"1"},…]}
{"ID":"AA356","Timestamp":"20140103", "Usefulness":"No",
"Code":[{"event1":"B","result":"0"},…]}
These are not comma-separated. The fundamental goal would be to parse certain fields (or all fields) into an R data.frame or data.table:
Timestamp Usefulness
0 20140101 Yes
1 20140102 No
2 20140103 No
Normally, I would read in a JSON within R as follows:
library(jsonlite)
jsonfile = "example1.json"
foobar = fromJSON(jsonfile)
This however throws a parsing error:
Error: lexical error: invalid char in json text.
[{"event1":"A","result":"1"},…]} {"ID":"1A35B","Timestamp"
(right here) ------^
This is a similar question to the following, but in R: multiple Json objects in one file extract by python
EDIT: This file format is called a "newline delimited JSON", NDJSON.
The three dots ... invalidate your JSON, hence your lexical error.
You can use jsonlite::stream_in() to 'stream in' lines of JSON.
library(jsonlite)
jsonlite::stream_in(file("~/Desktop/examples1.json"))
# opening file input connection.
# Imported 3 records. Simplifying...
# closing file input connection.
# ID Timestamp Usefulness Code
# 1 12345 20140101 Yes A, 1
# 2 1A35B 20140102 No B, 1
# 3 AA356 20140103 No B, 0
Data
I've cleaned your example data to make it valid JSON and saved it to my desktop as ~/Desktop/examples1.json
{"ID":"12345","Timestamp":"20140101", "Usefulness":"Yes","Code":[{"event1":"A","result":"1"}]}
{"ID":"1A35B","Timestamp":"20140102", "Usefulness":"No","Code":[{"event1":"B","result":"1"}]}
{"ID":"AA356","Timestamp":"20140103", "Usefulness":"No","Code":[{"event1":"B","result":"0"}]}
I have fixed width data files (.dbf) that don't have line separators. Here is what two lines of that datafile looks like:
20141101 77h 3.210 0 3 20141102 76h 3.090 0 3
The widths of one line is c(8,4,7,41) for date (8), some time measure (4), the data point (7), and some other columns that i can summarize in one "rest" column (41). After one line there is no separator and the next line is just appended to the first line. All time steps are basically written consecutively in one massive line. There is exclusively numbers, characters and white space in this file.
With read.fwf('filepath', widths = c(8,4,7,41)) R stops reading after the first line due to lack of line separator.
Is there an argument to tell read.fwf() when to start reading the new line when there is no line separator? Or should i use a different read command?
Thanks in advance.
Maybe not the best idea but this should work:
content <- scan('filepath','character',sep='~') # Warning choose a sep not appearing in datas to get the whole file.
# Split content in lines:
lines <- regmatches(content,gregexpr('.{60}',content))[[1]]
x <- tempfile()
write(lines,x)
data <- read.fwf(x, widths = c(8,4,7,41))
unlink(x)
The idea is to read the whole file, get each occurence of 60 chars into a single entry, write this to a tempfile, and read the data from this tempfile before deleting the temporary file.
Another approach is doable with regexes and package stringr (still with content resulting from scan above):
library(stringr)
d <- data.frame( str_match_all( content, "(.{8})(.{4})(.{7})(.{41})")[[1]][,2:5], stringsAsFactors=FALSE)
which gives:
V1 V2 V3 V4
1 20141101 77h 3.210 0 3
2 20141102 76h 3.090 0 3
str_match_all return a list, here with 1 element because there's only one line as input, so we remove it with [[1]].
Now the return is 5 columns, the first one being the full match, others being the capture groups so we subset the matrix on columns 2 to 5 to get only the 4 columns we need and wrap it in as.data.frame to get a data.frame at end.
you can then name the columns with colnames(d) <- c('date','time','data_point','rest')
If you wish to clean up the white spaces you can wrap the str_extract_all result in trimws (thanks to #jaap for the remind of this function) like this:
td <- data.frame( trimws( str_match_all( content, "(.{8})(.{4})(.{7})(.{41})")[[1]][,2:5] ), stringsAsFactors=FALSE)
Output:
X1 X2 X3 X4
1 20141101 77h 3.210 0 3
2 20141102 76h 3.090 0 3
A different, and probably less elegant, solution with readLines, substr, trimws, separate (tidyr) and mutate_all (dplyr):
txt <- readLines('filepath')
dfx <- data.frame(V1 = sapply(seq(from=1, to=nchar(txt), by=60),
function(x) substr(txt, x, x+59)))
library(dplyr)
library(tidyr)
dfx %>%
separate(V1, c(paste0("V",LETTERS[1:5])), c(8,12,19,55)) %>%
mutate_all(trimws)
which gives:
VA VB VC VD VE
1 20141101 77h 3.210 0 3
2 20141102 76h 3.090 0 3
To get different column names , just replace c(paste0("V",LETTERS[1:5]) with a vector of columnnames you want.
If you want to transform the columns into the correct classes instead of into character, you can use funs(ul = type.convert(trimws(.))) inside mutate_all.
In addition to the other answers, some general info about dbf files:
Unless this is a one time read of a static file, it would be best to check the file/fields structure first in case that changes over time. See here for the internal structure of a dbf file.
But maybe even more important:
Each record in a dbf file is preceded by one byte for the delete flag. If this is a space, the record is not deleted, if it's an asterisk * the record is marked for deletion (records are not removed from a dbf file until the file is packed), and you probably want to skip those records. The first part of the data could also be overwritten with "DELETED" for example.
So, in your record c(8,4,7,41), the last byte of the rest column (41) is actually the delete flag of the record that follows it - and the last record in the file will only have 40 bytes for that field (but if you're lucky, the file has an EOF marker (0x1a), so maybe you didn't have a problem with the size there).
Thus, your record should actually be: c(1,8,4,7,40), where the 1 is the delete flag, and starting one byte sooner.
I am reading in a file of data that looks like this:
userId, fullName,email,password,activated,registrationDate,locale,notifyOnUpdates,lastSyncTime,plan_id,plan_period_months,plan_price,plan_exp_date,plan_is_trial,plan_is_trial_used,q_hear,q_occupation,pp_subid,pp_payments,pp_since,pp_cancelled,apikey
"2","John Smith,"john.smith#gmail.com","a","1","2004-07-23 14:19:32","en_US","1","2011-04-07 07:29:17","3",\N,\N,\N,"0","1",\N,\N,\N,\N,\N,\N,"d7734dce-4ae2-102a-8951-0040ca38ff83"
but the actual file as around 20000 records. I use the following R code to read it in:
user = read.csv("~/Desktop/dbdump/users.txt", na.strings = "\\N", quote="")
And the reason I have quote="" is because without it the import stops prematurely. I end up with a total of 9569 observations. Why I don't understand why exactly the quote="" overcomes this problem, it seems to do so.
Except that it introduces other problems that I have to 'fix'. The first one I saw is that the dates end up being strings which include the quotes, which don't want to convert to actual dates when I use to.Date() on them.
Now I could fix the strings and hack my way through. But better to know more about what I am doing. Can someone explain:
Why does the quote="" fix the 'bad data'
What is a best-practice technique to figure out what is causing the read.csv to stop early? (If I just look at the input data at +/- the indicated row, I don't see anything amiss).
Here are the lines 'near' the 'problem'. I don't see the damage do you?
"16888","user1","user1#gmail.com","TeilS12","1","2008-01-19 08:47:45","en_US","0","2008-02-23 16:51:53","1",\N,\N,\N,"0","0","article","student",\N,\N,\N,\N,"ad949a8e-17ed-102b-9237-0040ca390025"
"16889","user2","user2#gmail.com","Gaspar","1","2008-01-19 10:34:11","en_US","1",\N,"1",\N,\N,\N,"0","0","email","journalist",\N,\N,\N,\N,"8b90f63a-17fc-102b-9237-0040ca390025"
"16890","user3","user3#gmail.com","boomblaadje","1","2008-01-19 14:36:54","en_US","0",\N,"1",\N,\N,\N,"0","0","article","student",\N,\N,\N,\N,"73f31f4a-181e-102b-9237-0040ca390025"
"16891","user4","user4#gmail.com","mytyty","1","2008-01-19 15:10:45","en_US","1","2008-01-19 15:16:45","1",\N,\N,\N,"0","0","google-ad","student",\N,\N,\N,\N,"2e48e308-1823-102b-9237-0040ca390025"
"16892","user5","user5#gmail.com","08091969","1","2008-01-19 15:12:50","en_US","1",\N,"1",\N,\N,\N,"0","0","dont","dont",\N,\N,\N,\N,"79051bc8-1823-102b-9237-0040ca390025"
* Update *
It's more tricky. Even though the total number of rows imported is 9569, if I look at the last few rows they correspond to the last few rows of data. Therefore I surmise that something happened during the import to cause a lot of rows to be skipped. In fact 15914 - 9569 = 6345 records. When I have the quote="" in there I get 15914.
So my question can be modified: Is there a way to get read.csv to report about rows it decides not to import?
* UPDATE 2 *
#Dwin, I had to remove na.strings="\N" because the count.fields function doesn't permit it. With that, I get this output which looks interesting but I don't understand it.
3 4 22 23 24
1 83 15466 178 4
Your second command produces a lots of data (and stops when max.print is reached.) But the first row is this:
[1] 2 4 2 3 5 3 3 3 5 3 3 3 2 3 4 2 3 2 2 3 2 2 4 2 4 3 5 4 3 4 3 3 3 3 3 2 4
Which I don't understand if the output is supposed to show how many fields there are in each record of input. Clearly the first lines all have more than 2,4,2 etc fields... Feel like I am getting closer, but still confused!
The count.fields function can be very useful in identifying where to look for malformed data.
This gives a tabulation of fields per line ignores quoting, possibly a problem if there are embedded commas:
table( count.fields("~/Desktop/dbdump/users.txt", quote="", sep=",") )
This give a tabulation ignoring both quotes and "#"(octothorpe) as a comment character:
table( count.fields("~/Desktop/dbdump/users.txt", quote="", comment.char="") )
Atfer seeing what you report for the first tabulation..... most of which were as desired ... You can get a list of the line positions with non-22 values (using the comma and non-quote settings):
which( count.fields("~/Desktop/dbdump/users.txt", quote="", sep=",") != 22)
Sometimes the problem can be solved with fill=TRUE if the only difficulty is missing commas at the ends of lines.
One problem I have spotted (thanks to data.table) is the missing quote (") after John Smith. Could this be a problem also for other lines you have?
If I add the "missing" quote after John Smith, it reads fine.
I saved this data to data.txt:
userId, fullName,email,password,activated,registrationDate,locale,notifyOnUpdates,lastSyncTime,plan_id,plan_period_months,plan_price,plan_exp_date,plan_is_trial,plan_is_trial_used,q_hear,q_occupation,pp_subid,pp_payments,pp_since,pp_cancelled,apikey
"2","John Smith","john.smith#gmail.com","a","1","2004-07-23 14:19:32","en_US","1","2011-04-07 07:29:17","3",\N,\N,\N,"0","1",\N,\N,\N,\N,\N,\N,"d7734dce-4ae2-102a-8951-0040ca38ff83"
"16888","user1","user1#gmail.com","TeilS12","1","2008-01-19 08:47:45","en_US","0","2008-02-23 16:51:53","1",\N,\N,\N,"0","0","article","student",\N,\N,\N,\N,"ad949a8e-17ed-102b-9237-0040ca390025"
"16889","user2","user2#gmail.com","Gaspar","1","2008-01-19 10:34:11","en_US","1",\N,"1",\N,\N,\N,"0","0","email","journalist",\N,\N,\N,\N,"8b90f63a-17fc-102b-9237-0040ca390025"
"16890","user3","user3#gmail.com","boomblaadje","1","2008-01-19 14:36:54","en_US","0",\N,"1",\N,\N,\N,"0","0","article","student",\N,\N,\N,\N,"73f31f4a-181e-102b-9237-0040ca390025"
"16891","user4","user4#gmail.com","mytyty","1","2008-01-19 15:10:45","en_US","1","2008-01-19 15:16:45","1",\N,\N,\N,"0","0","google-ad","student",\N,\N,\N,\N,"2e48e308-1823-102b-9237-0040ca390025"
"16892","user5","user5#gmail.com","08091969","1","2008-01-19 15:12:50","en_US","1",\N,"1",\N,\N,\N,"0","0","dont","dont",\N,\N,\N,\N,"79051bc8-1823-102b-9237-0040ca390025"
And this is a code. Both fread and read.csv works fine.
require(data.table)
dat1 <- fread("data.txt", header = T, na.strings = "\\N")
dat1
dat2 <- read.csv("data.txt", header = T, na.strings = "\\N")
dat2