reading .csv file with decimals separated by a comma with CSV.jl - csv

Question

I am trying to read some data into julia into a data frame to work with it. A minimal example of the .csv file could look like this:
A; B; C; D
ab; 1,23; 4; 9,2
ab; 3,4; 7; 1,1
ba; 6; 2,3; 8,6
I load the following to packages and read the data:
using DataFrames
using CSV
d = CSV.read( "test.csv", delim=";")
Julia recognizes the following types:
eltypes(d)
CategoricalArrays.CategoricalString{UInt32}
String
String
String
How could I now turn whole columns to floats with the comma replaced by a dot? My first idea was to use:
float(d[1,2])
But I did not find an option to tell julia to replace the comma with a dot.
My next idea was to first replace the comma and then convert it:
float(replace(d[1,2], ",", "."))
That works fine on a single cell but not on a whole column:
float(replace(d[:,2], ",", "."))
MethodError: no method matching
replace(::WeakRefStrings.WeakRefStringArray{WeakRefString{UInt8},1,Union{}},
::String, ::String)
I also tried:
d = CSV.read( "test.csv", delim=";", decimal=",")
which also just gives an error ...
Any ideas how to handle this problem and how to efficiently read the data into julia?
Thanks a lot!
Best regards.

Answer 1

One straightforward way is to read the file to string, replace the comma decimal separators by dots and then create the DataFrame from it:
s = replace(readstring("test.csv"), ",", ".")
CSV.read(IOBuffer(s); delim=';', types=[String, Float64, Float64, Float64])
Note that you can use the types keyword to specifiy the column types (it will then implicitly parse the string entries).
EDIT: According to this github issue the CSV.jl's read method supports a decimal keyword (from version v0.2.0 on) which allows you to do
CSV.read("test.csv"; delim=';', decimal=',', types=[String, Float64, Float64, Float64])
EDIT: Removed hint to alternatively use readtable from DataFrames.jl because it seems to be deprecated in favor of CSV.read.