Reading accuracy
Yuri Belenky
When reading from file with the "mfscanf(-1, fd, "%e\n")" function
the numbers "-101061877.000000" and "970551467.000000" represented by
"- 1.011D+08" and "9.706D+08". When I trying to divide them for example
by 1000000 the results are "- 101.06188" and "970.55149" respectively.
How to increase accuracy of reading?
Thanks a lot,
Y.B.
Francois Vogel
lost in this process if the correct format parameters are sent to mfscanf.
I believe your problem is just a display problem.
See help format.
Example:
-->a=-101061877.000000
a =
- 1.011E+08
-->a/1000000
ans =
- 101.06188
-->format("e",15)
-->a/1000000
ans =
- 1.01061877E+02
HTH
Francois
"Yuri Belenky" <n...@spam.net> a écrit dans le message de news:
dk59qg$doi$1...@domitilla.aioe.org...
Gaspard Lequeux
Hej,
A clumsy work around might be to read al your data in as strings and then use
execstr to convert them to numbers:
-->a=msscanf("970551467.00000","%e")
a =
9.706E+08
-->b=a/1000000
b =
970.55149
-->execstr("a=970551467.00000")
-->b=a/1000000
b =
970.55147
-->
And maybe submit a bug report...
I have some functions who implement that (needed that because of speed issues
and a (submitted) bug that crashed scilab on MSwindows on certain big numbers).
It's not a replacement for mfscanf through. You will have to do some work
yourself. If you want them, ask.
wfg,
Gaspard
Gaspard Lequeux
On Mon, 31 Oct 2005, Francois Vogel wrote:
> Reading from file should be in the correct format, I mean, no precision is
> lost in this process if the correct format parameters are sent to mfscanf.
>
> I believe your problem is just a display problem.
Apparently not:
-->a=msscanf("970551467.00000","%e")
a =
9.706E+08
-->a/1000000
ans =
970.55149
-->format("e",15)
-->a/1000000
ans =
9.70551488E+02
-->a=msscanf("970551467.00000","%15e")
a =
9.70551488E+08
Or is there something I miss? I must admit it is on scilab 3.0...
wfg,
Gaspard
Francois Vogel
How interesting... Same result on my very recent cvs version.
Could be bug 1315, or maybe 583.
Francois
Enrico Segre
> Reading from file should be in the correct format, I mean, no precision is
> lost in this process if the correct format parameters are sent to mfscanf.
>
> I believe your problem is just a display problem.
>
> See help format.
No, the OP is correct.
-->format(20)
-->fd=mopen("a.dat");b=mfscanf(-1, fd, "%e\n"),mclose(fd);
b =
! - 101061880. !
! 970551488. !
-->fd=mopen("a.dat");b=mfscanf(-1, fd, "%f\n"),mclose(fd);
b =
! - 101061880. !
! 970551488. !
-->fd=mopen("a.dat");b=mfscanf(-1, fd, "%g\n"),mclose(fd);
b =
! - 101061880. !
! 970551488. !
however note that:
-->fd=mopen("a.dat");b=mfscanf(-1, fd, "%d\n"),mclose(fd);
b =
- 101061877.
-->fd=mopen("a.dat");b=mfscanf(-1, fd, "%d.\n"),mclose(fd); b =
! - 101061877. !
! 0. !
! 970551467. !
! 0. !
And
-->read("a.dat",-1,1)
ans =
! - 101061877. !
! 970551467. !
Pehaps mfscanf implements a float read instead of a double?
Enrico
Yuri Belenky
Yuri.