Adrian--thanks!
I have a followup: parse seems about 100 times slower than the equivalent
C++ code. I would have hoped for only a
factor of two or so. Here is a C++ code fragment:
for(int i = 1; i <= 1000000; i++)
{
std::string s("4.123456789");
std::istringstream iss(s);
double d;
iss >> d;
}
This takes 0.8 seconds to run on a mac mini, so about a microsecond per
parsing of a double.
Here is a julia version, with only 10^4 parses:
function testparse()
parse("4.123456789")
end
@time testparse()
@time for i=1:10000
testparse()
end
Running this on the same machine takes about 0.65 seconds. I get similar
results if I call @time for a single call to
testparse(), and any other variations I could think of. This indicates
julia's parse is about a factor of 80 slower than the equivalent
in C++.
Have I done something to make this slow, or is parse just very slow? Why
would it be so slow?
-Steve
On Friday, December 25, 2015 at 2:20:21 AM UTC-8, Adrian Cuthbertson wrote:
>
> The essential part first:
>
> julia> a,b,c,d=map(parse,split("9 2 2 1.3\n"))
> 4-element Array{Real,1}:
> 9
> 2
> 2
> 1.3
>
> but see what you actually get back...
> julia> typeof(a)
> Int64
> julia> typeof(d)
> Float64
>
> That's Julia's type inferencing at play. And functional abilities (map,
> parse). See the manual on those.
>
> Now as originally needed...
>
> julia> io=IOBuffer("9 2 2 1.3\n")
> IOBuffer(data=UInt8[...], readable=true, writable=false, seekable=true,
> append=false, size=10, maxsize=Inf, ptr=1, mark=-1)
>
> julia> n,nup,ndn,a=map(parse,split(readline(io)))
> 4-element Array{Real,1}:
> 9
> 2
> 2
> 1.3
>
> julia> typeof(n)
> Int64
>
> julia> typeof(a)
> Float64
>
> It's probably better to do it properly in a function rather than like that
> as a one-liner - so you can catch exceptions, etc.
>
> -- Adrian.
>
>
> On Thu, Dec 24, 2015 at 8:36 PM, Steven White <[email protected]
> <javascript:>> wrote:
>
>> Say I have a data file in plain text, which starts with some lines that
>> describe matrix sizes, etc, for the data that follows. For example, the
>> first line might be
>> 9 2 2 1.3
>>
>> To read this file, in C I would use something like this call to scanf:
>> scanf("%d%d%d%lf",&n,&nup,&ndn,&a);
>>
>> How do I do this in julia? There doesn't seem to be a simple, obvious
>> way. Of course, you can read the line as a string, and parse it using
>> regular expressions, but this is much messier than the call to scanf.
>>
>> Here is a solution:
>> getline() = readdlm(IOBuffer(readline()),Any)
>>
>> n,nup,ndn,a = getline()
>>
>> This seems to work reliably. (For example, it doesn't choke if there
>> happen to be two spaces between some of the numbers.)
>>
>> It is also obscure and doesn't give any warnings if some of the types
>> aren't what was expected. It also seems inefficient if you try to read a
>> lot of data this way.
>>
>> Questions: Why isn't something like this (but better) built in to
>> julia? Is there some nice method that I've missed that does this better?
>> Can this getline() function be improved?
>>
>>
>
