[ncl-talk] Floating point issue

Wed Feb 6 19:46:18 MST 2019

Thank you for explain it, Dave! It works!

Graziela.

Em qua, 6 de fev de 2019 às 17:14, Dave Allured - NOAA Affiliate <
dave.allured at noaa.gov> escreveu:

> Graziela,
>
> This is a normal issue with floating point numbers.  Decimal fractions can
> not be represented exactly in binary floating point, except for a small
> subset that are exact sums of negative powers of 2.  For example, 42.75 can
> be represented exactly, but 42.6 can not.  When reading decimal strings
> from text formats like CSV, NCL and other languages will generally select
> the nearest available floating point value.  Therefore, when working with
> assumed decimal numbers, you get small roundoff errors that need special
> treatment.  Please see this article:
>
>
> https://en.wikipedia.org/wiki/Floating-point_arithmetic#Representable_numbers,_conversion_and_rounding
>
> NCL's default print function does not show all significant floating point
> digits.  To properly debug this, use sprintf to show all significant
> digits.  Single precision float requires at least 9 significant decimal
> digits to display all possible unique internal values.  Double precision
> requires at least 17 decimal digits to show all possible values uniquely.
>
>     print (sprintf ("%27.17g", (/ a, b, c /) ))
>
> The normal way to handle inexact comparisons is to set a small threshold
> based on the expected numeric range and precision, then use threshold
> comparisons rather than equality.  Threshold comparisons may be absolute or
> relative, and should be customized for each application.  Try this:
>
>     delta = 0.001
>     test1 = -50.2
>     b = ind ( abs (xlon1(0:29) - test1) .lt. delta )
>
> Finally, single precision float is okay for one or two fractional digits,
> but is often not precise enough for geographic coordinates in many
> applications.  I recommend you upgrade to type double for all processing of
> lat/lon coordinates.  When reading CSV, use todouble rather than tofloat.
> Also decrease the threshold value "delta"  in threshold comparisons.
>
> --Dave
>
>
> On Wed, Feb 6, 2019 at 6:31 AM Graziela Luzia <grazi.luzia at gmail.com>
> wrote:
> >
> > Hello ncl-users.
> >
> > I'm converting a csv file in netcdf and apparently my script, although
> not very efficient (I'm a begginer) was working fine. But for some points
> it dindn't write correctly. I've checked everything and I supose the
> problem is converting string to float, but I can't figure out how can I
> solve this problem.
> >
> > I don't understand why it shows me a value like 42.6 but it is like it
> was 42.600002. It causes me trouble when I compare it with another variable
> equal to 42.6, it returns a false result. Also, it only happen to some
> values, like 42.6 (but not with 50.2) as you can see below.
> >
> > I've tried to use decimalPlaces, but it results the same.
> > I'd apreciate any information. Thanks in advance.
> >
> > uname-a: Linux Tower2 4.15.0-45-generic #48~16.04.1-Ubuntu SMP Tue Jan
> 29 18:03:48 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux
> >
> > Graziela.
> >
> > ---
> > lines = asciiread("resultado.csv",-1,"string")
> > data = lines(1:) ; exclude line 0 header
> >
> > ;xlon = decimalPlaces(tofloat(str_get_field(data,1,",")),1,True)
> > xlon=tofloat(str_get_field(data,1,","))
> >
> > xlon1 = new(600,"float",-999)
> >    do i=0,599
> >       xlon1(i)=xlon(i*8)+0.1
> >    end do
> >
> > print(xlon1(0:29))
> >
> > a=xlon1(23)-xlon1(24)
> > print(a)
> >
> > b=ind(xlon1(0:29).eq.-50.2)
> > c=ind(xlon1(0:29).eq.-42.6)
> > print(b)
> > print(c)
> >
> > ----
> > labren at Tower2:~/Documentos/Produto3$ ncl teste.ncl
> >  Copyright (C) 1995-2015 - All Rights Reserved
> >  University Corporation for Atmospheric Research
> >  NCAR Command Language Version 6.3.0
> >  The use of this software is governed by a License Agreement.
> >  See http://www.ncl.ucar.edu/ for more details.
> >
> > Variable: xlon1 (subsection)
> > Type: float
> > Total Size: 120 bytes
> >             30 values
> > Number of Dimensions: 1
> > Dimensions and sizes: [30]
> > Coordinates:
> > Number Of Attributes: 1
> >   _FillValue : -999
> > (0) -50.2
> > (1) -50.2
> > (2) -50.2
> > (3) -50.2
> > (4) -50.2
> > (5) -50.2
> > (6) -50.2
> > (7) -50.2
> > (8) -50.2
> > (9) -50.2
> > (10) -50.2
> > (11) -50.2
> > (12) -50.2
> > (13) -50.2
> > (14) -50.2
> > (15) -50.2
> > (16) -50.2
> > (17) -50.2
> > (18) -50.2
> > (19) -50.2
> > (20) -50.2
> > (21) -50.2
> > (22) -50.2
> > (23) -50.2
> > (24) -42.6
> > (25) -42.6
> > (26) -42.6
> > (27) -42.6
> > (28) -42.6
> > (29) -42.6
> >
> > Variable: a
> > Type: float
> > Total Size: 4 bytes
> >             1 values
> > Number of Dimensions: 1
> > Dimensions and sizes: [1]
> > Coordinates:
> > Number Of Attributes: 1
> >   _FillValue : -999
> > (0) -7.599998
> >
> >
> > Variable: b
> > Type: integer
> > Total Size: 96 bytes
> >             24 values
> > Number of Dimensions: 1
> > Dimensions and sizes: [24]
> > Coordinates:
> > (0) 0
> > (1) 1
> > (2) 2
> > (3) 3
> > (4) 4
> > (5) 5
> > (6) 6
> > (7) 7
> > (8) 8
> > (9) 9
> > (10) 10
> > (11) 11
> > (12) 12
> > (13) 13
> > (14) 14
> > (15) 15
> > (16) 16
> > (17) 17
> > (18) 18
> > (19) 19
> > (20) 20
> > (21) 21
> > (22) 22
> > (23) 23
> >
> > Variable: c
> > Type: integer
> > Total Size: 4 bytes
> >             1 values
> > Number of Dimensions: 1
> > Dimensions and sizes: [1]
> > Coordinates:
> > Number Of Attributes: 1
> >   _FillValue : -2147483647
> > (0) -2147483647
> >
> > My csv file contain values like this:
> >
> > xlon,ylat,ano,mes,direcao,v1,v2,v3,v4
> > -50.3,-30.4,2016,1,45,0,0.0020206556,0.0017961383
> > -50.3,-30.4,2016,1,90,0.0002245173,0.0004490346,0.0049393803,0.0246969017
> >
> -50.3,-30.4,2016,1,135,0.0011225864,0.0029187247,0.0114503817,0.0177368657
> > -50.3,-30.4,2016,1,180,0,0.0017961383,0.0038167939,0.0141445891 (...)
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20190207/435b7f25/attachment.html>