[R] cut{base}: is it a bug?
David Winsemius
dw|n@em|u@ @end|ng |rom comc@@t@net
Mon Sep 24 22:12:34 CEST 2018
> On Sep 24, 2018, at 12:00 PM, Jose Claudio Faria <joseclaudio.faria using gmail.com> wrote:
>
> Dears,
>
> Thank you for your contribution!
>
> However, this function is important in a generic usage package for
> frequency distribution tables: fdth (
> https://cran.r-project.org/web/packages/fdth/index.html).
>
> In this case, when I do not know in advance what the user data is, what is
> the best option to avoid deviations as centuados as the example?
>
> The data used in the example was sent to me from a teacher trying to
> reproduce in class the table of a book.
If you want to provide tools that protect unsuspecting users from falling into common numerical and well understood potential traps, then why don't you process your data inputs with round( obj, 8) or something similar to your liking.
> round( 3*.1, 8) == 3*.1
[1] FALSE
> 0.3 == 3*.1
[1] FALSE
> round( 3*.1, 8) == 0.3
[1] TRUE
--
David.
>
> Best,
>
> ///\\\///\\\///\\\///\\\///\\\///\\\///\\\///\\\
> Jose Claudio Faria
> UESC/DCET/Brasil
> joseclaudio.faria at gmail.com
> Telefones:
> 55(73)3680.5545 - UESC
> 55(73)99966.9100 - VIVO
> 55(73)98817.6159 - OI
> ///\\\///\\\///\\\///\\\///\\\///\\\///\\\///\\\
>
> If you have software to deal with statistics, you have arms;
> if you have good software, you have arms and legs;
> if you have software like R, you have arms, legs and wings...
> the height of your flight depends only on you!
>
> 2018-09-24 14:42 GMT-03:00 David L Carlson <dcarlson using tamu.edu>:
>
>> Yes, I should have included that point. The cut() function "encourages"
>> exact comparison of values by including the right= argument without a
>> warning that this may create unexpected results. With truly continuous
>> data, values falling exactly on the boundary would be rare.
>>
>> Most data arrives from instruments that measure to limited precision.
>> Introductory statistics texts deal with this by distinguishing between
>> "true" and "stated" class limits. Or, like Lyman Ott, recommend choosing
>> the starting point interval such that "no measurement falls on a point of
>> division between two intervals."
>>
>> ----------------------------------------
>> David L Carlson
>> Department of Anthropology
>> Texas A&M University
>> College Station, TX 77843-4352
>>
>> -----Original Message-----
>> From: Jeff Newmiller <jdnewmil using dcn.davis.ca.us>
>> Sent: Monday, September 24, 2018 10:41 AM
>> To: r-help using r-project.org; David L Carlson <dcarlson using tamu.edu>; Jose
>> Claudio Faria <joseclaudio.faria using gmail.com>; r-help using r-project.org
>> Subject: Re: [R] cut{base}: is it a bug?
>>
>> "Subtracting a bit" only fixes the problem for the test data... it
>> introduces a bias in any continuous data you happen to throw at it.
>> However, if you have data with known rounding applied (e.g. published
>> tabular data) then the subtracting trick can be useful. In general you
>> should not expect floating point fractions to behave like exact values in
>> your analysis.
>>
>> On September 24, 2018 8:14:09 AM PDT, David L Carlson <dcarlson using tamu.edu>
>> wrote:
>>> You've been bitten by FAQ 7.31: Why doesn't R think these numbers are
>>> equal?
>>> https://cran.r-project.org/doc/FAQ/R-FAQ.html#Why-doesn_
>> 0027t-R-think-these-numbers-are-equal_003f
>>>
>>> Your boundaries and your data values are not what you think they are.
>>> This is a limitation of digital computing not R.
>>>
>>>> print(seq(from=.6, to=2.2, by=.2), digits=17)
>>> [1] 0.59999999999999998 0.80000000000000004 1.00000000000000000
>>> 1.20000000000000018
>>> [5] 1.39999999999999991 1.60000000000000009 1.80000000000000027
>>> 2.00000000000000000
>>> [9] 2.20000000000000018
>>>
>>>> print(dat, digits=17)
>>> [1] 0.59999999999999998 0.59999999999999998 0.59999999999999998
>>> 0.69999999999999996
>>> [5] 0.69999999999999996 0.69999999999999996 0.69999999999999996
>>> 0.69999999999999996
>>> [9] 0.80000000000000004 0.80000000000000004 0.80000000000000004
>>> 0.90000000000000002
>>> [13] 0.90000000000000002 0.90000000000000002 0.90000000000000002
>>> 1.00000000000000000
>>> [17] 1.00000000000000000 1.00000000000000000 1.00000000000000000
>>> 1.10000000000000009
>>> [21] 1.10000000000000009 1.10000000000000009 1.19999999999999996
>>> 1.19999999999999996
>>> [25] 1.19999999999999996 1.19999999999999996 1.30000000000000004
>>> 1.30000000000000004
>>> [29] 1.30000000000000004 1.39999999999999991 1.39999999999999991
>>> 1.39999999999999991
>>> [33] 1.50000000000000000 1.50000000000000000 1.50000000000000000
>>> 1.60000000000000009
>>> [37] 1.60000000000000009 1.69999999999999996 1.69999999999999996
>>> 1.69999999999999996
>>> [41] 1.69999999999999996 1.80000000000000004 1.80000000000000004
>>> 1.80000000000000004
>>> [45] 1.89999999999999991 1.89999999999999991 2.00000000000000000
>>> 2.00000000000000000
>>> [49] 2.00000000000000000 2.00000000000000000 2.00000000000000000
>>> 2.10000000000000009
>>>
>>> The simplest solution is to subtract a bit. This also means you don't
>>> need the include.lowest= or right= arguments:
>>>
>>>> f <- cut(dat,
>>> + breaks= seq(from=.6-.01, to=2.2-.01, by=.2),
>>> + dig.lab=10L)
>>>> as.matrix(tb <- table(f))
>>> [,1]
>>> [0.59,0.79) 8
>>> [0.79,0.99) 7
>>> [0.99,1.19) 7
>>> [1.19,1.39) 7
>>> [1.39,1.59) 6
>>> [1.59,1.79) 6
>>> [1.79,1.99) 5
>>> [1.99,2.19] 6
>>>
>>> ----------------------------------------
>>> David L Carlson
>>> Department of Anthropology
>>> Texas A&M University
>>> College Station, TX 77843-4352
>>>
>>>
>>> -----Original Message-----
>>> From: R-help <r-help-bounces using r-project.org> On Behalf Of Jose Claudio
>>> Faria
>>> Sent: Monday, September 24, 2018 9:32 AM
>>> To: r-help using r-project.org
>>> Subject: [R] cut{base}: is it a bug?
>>>
>>> Dears members,
>>>
>>> Is the below a bug of the cut {base} function?
>>>
>>> dat <- c(
>>> 0.6, 0.6, 0.6, 0.7, 0.7, 0.7, 0.7, 0.7, #(8)
>>> 0.8, 0.8, 0.8, 0.9, 0.9, 0.9, 0.9, #(7)
>>> 1.0, 1.0, 1.0, 1.0, 1.1, 1.1, 1.1, #(7)
>>> 1.2, 1.2, 1.2, 1.2, 1.3, 1.3, 1.3, #(7)
>>> 1.4, 1.4, 1.4, 1.5, 1.5, 1.5, #(6)
>>> 1.6, 1.6, 1.7, 1.7, 1.7, 1.7, #(6)
>>> 1.8, 1.8, 1.8, 1.9, 1.9, #(5)
>>> 2.0, 2.0, 2.0, 2.0, 2.0, 2.1 #(6)
>>> )
>>>
>>> # making class from function "cut"
>>> (f <- cut(dat,
>>> breaks= seq(from=.6, to=2.2, by=.2),
>>> include.lowest=TRUE,
>>> dig.lab=10L,
>>> right=FALSE))
>>>
>>> # more easy to see the table
>>> as.matrix(tb <- table(f))
>>>
>>> # Checking
>>> print(length(dat[dat >= 0.6 & dat < 0.8])) == tb[1]
>>> print(length(dat[dat >= 0.8 & dat < 1.0])) == tb[2]
>>> print(length(dat[dat >= 1.0 & dat < 1.2])) == tb[3] # !?
>>> print(length(dat[dat >= 1.2 & dat < 1.4])) == tb[4] # !?
>>> print(length(dat[dat >= 1.4 & dat < 1.6])) == tb[5]
>>> print(length(dat[dat >= 1.6 & dat < 1.8])) == tb[6] # !?
>>> print(length(dat[dat >= 1.8 & dat < 2.0])) == tb[7] # !?
>>> print(length(dat[dat >= 2.0 & dat < 2.2])) == tb[8]
>>>
>>> Best,
>>> ///\\\///\\\///\\\///\\\///\\\///\\\///\\\///\\\
>>> Jose Claudio Faria
>>> UESC/DCET/Brasil
>>> joseclaudio.faria at gmail.com
>>> Telefones:
>>> 55(73)3680.5545 - UESC
>>> 55(73)99966.9100 - VIVO
>>> 55(73)98817.6159 - OI
>>> ///\\\///\\\///\\\///\\\///\\\///\\\///\\\///\\\
>>>
>>> If you have software to deal with statistics, you have arms; if you
>>> have good software, you have arms and legs; if you have software like
>>> R, you have arms, legs and wings...
>>> the height of your flight depends only on you!
>>>
>>> [[alternative HTML version deleted]]
>>>
>>> ______________________________________________
>>> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>> PLEASE do read the posting guide
>>> http://www.R-project.org/posting-guide.html
>>> and provide commented, minimal, self-contained, reproducible code.
>>>
>>> ______________________________________________
>>> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>> PLEASE do read the posting guide
>>> http://www.R-project.org/posting-guide.html
>>> and provide commented, minimal, self-contained, reproducible code.
>>
>> --
>> Sent from my phone. Please excuse my brevity.
>>
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help using r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
David Winsemius
Alameda, CA, USA
'Any technology distinguishable from magic is insufficiently advanced.' -Gehm's Corollary to Clarke's Third Law
More information about the R-help
mailing list