How to round to nearest multiple? - julia

Is there an idiomatic way to round to the nearest multiple of a number?
number multiple result
12.2 0.5 12.0
12.2 0.25 12.25
12.4 0.5 12.5

You can define a function:
round_step(x, step) = round(x / step) * step
Usage:
julia> round_step(12.2, 0.25)
12.25
Such a function is actually used internally in Base for rounding numbers to a certain number of digits in a certain base:
julia> Base._round_step(12.2, 0.25, RoundNearest)
12.25
However since this is an internal implementation detail you shouldn't rely on this function. This function in turn calls _round_invstep.
julia> Base._round_invstep(12.2, 4, RoundNearest)
12.25
julia> Base._round_invstep(12.4, 2, RoundNearest)
12.5
Which performs the operation round(x * invstep, r) / invstep.
Because your examples happen to correspond to 0.1 in base 4 and base 2 you can also use round directly for these particular cases:
julia> round(12.2; base=2, digits=1)
12.0
julia> round(12.2; base=4, digits=1)
12.25
julia> round(12.4; base=2, digits=1)
12.5

Related

How does R (v3.6.1) determine rounding with odd floor values? [duplicate]

Yes I know why we always round to the nearest even number if we are in the exact middle (i.e. 2.5 becomes 2) of two numbers. But when I want to evaluate data for some people they don't want this behaviour. What is the simplest method to get this:
x <- seq(0.5,9.5,by=1)
round(x)
to be 1,2,3,...,10 and not 0,2,2,4,4,...,10.
Edit: To clearify: 1.4999 should be 1 after rounding. (I thought this would be obvious)
This is not my own function, and unfortunately, I can't find where I got it at the moment (originally found as an anonymous comment at the Statistically Significant blog), but it should help with what you need.
round2 = function(x, digits) {
posneg = sign(x)
z = abs(x)*10^digits
z = z + 0.5 + sqrt(.Machine$double.eps)
z = trunc(z)
z = z/10^digits
z*posneg
}
x is the object you want to round, and digits is the number of digits you are rounding to.
An Example
x = c(1.85, 1.54, 1.65, 1.85, 1.84)
round(x, 1)
# [1] 1.8 1.5 1.6 1.8 1.8
round2(x, 1)
# [1] 1.9 1.5 1.7 1.9 1.8
(Thanks #Gregor for the addition of + sqrt(.Machine$double.eps).)
If you want something that behaves exactly like round except for those xxx.5 values, try this:
x <- seq(0, 1, 0.1)
x
# [1] 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
floor(0.5 + x)
# [1] 0 0 0 0 0 1 1 1 1 1 1
As #CarlWitthoft said in the comments, this is the IEC 60559 standard as mentioned in ?round:
Note that for rounding off a 5, the IEC 60559 standard is expected to be used, ‘go to the even digit’. Therefore round(0.5) is 0 and round(-1.5) is -2. However, this is dependent on OS services and on representation error (since e.g. 0.15 is not represented exactly, the rounding rule applies to the represented number and not to the printed number, and so round(0.15, 1) could be either 0.1 or 0.2).
An additional explanation by Greg Snow:
The logic behind the round to even rule is that we are trying to
represent an underlying continuous value and if x comes from a truly
continuous distribution, then the probability that x==2.5 is 0 and the
2.5 was probably already rounded once from any values between 2.45 and 2.54999999999999..., if we use the round up on 0.5 rule that we learned in grade school, then the double rounding means that values
between 2.45 and 2.50 will all round to 3 (having been rounded first
to 2.5). This will tend to bias estimates upwards. To remove the
bias we need to either go back to before the rounding to 2.5 (which is
often impossible to impractical), or just round up half the time and
round down half the time (or better would be to round proportional to
how likely we are to see values below or above 2.5 rounded to 2.5, but
that will be close to 50/50 for most underlying distributions). The
stochastic approach would be to have the round function randomly
choose which way to round, but deterministic types are not
comforatable with that, so "round to even" was chosen (round to odd
should work about the same) as a consistent rule that rounds up and
down about 50/50.
If you are dealing with data where 2.5 is likely to represent an exact
value (money for example), then you may do better by multiplying all
values by 10 or 100 and working in integers, then converting back only
for the final printing. Note that 2.50000001 rounds to 3, so if you
keep more digits of accuracy until the final printing, then rounding
will go in the expected direction, or you can add 0.000000001 (or
other small number) to your values just before rounding, but that can
bias your estimates upwards.
This appears to work:
rnd <- function(x) trunc(x+sign(x)*0.5)
Ananda Mahto's response seems to do this and more - I am not sure what the extra code in his response is accounting for; or, in other words, I can't figure out how to break the rnd() function defined above.
Example:
seq(-2, 2, by=0.5)
# [1] -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
round(x)
# [1] -2 -2 -1 0 0 0 1 2 2
rnd(x)
# [1] -2 -2 -1 -1 0 1 1 2 2
Depending on how comfortable you are with jiggling your data, this works:
round(x+10*.Machine$double.eps)
# [1] 1 2 3 4 5 6 7 8 9 10
This method:
round2 = function(x, n) {
posneg = sign(x)
z = abs(x)*10^n
z = z + 0.5
z = trunc(z)
z = z/10^n
z*posneg
}
does not seem to work well when we have numbers with many digits. E.g. doing round2(2436.845, 2) will give us 2436.84. The issue seems to occur with the trunc(z) function.
Overall, I think it has something to do with the way R stores numbers and thus the trunc and float function doesn't always work. I was able to get around it in not the most elegant way:
round2 = function(x, n) {
posneg = sign(x)
z = abs(x)*10^n
z = z + 0.5
z = trunc(as.numeric(as.character(z)))
z = z/10^n
(z)*posneg
}
This mimics the rounding away from zero at .5:
round_2 <- function(x, digits = 0) {
x = x + abs(x) * sign(x) * .Machine$double.eps
round(x, digits = digits)
}
round_2(.5 + -2:4)
-2 -1 1 2 3 4 5

Julia: How can we compute the adjoint or classical adjoint (Linear Algebra)?

I wan to compute the classical adjoint in Julia 1.0
For this I copied the matrix given as an example in wikipedia
julia> B = [-3 2 -5; -1 0 -2; 3 -4 1]
3×3 Array{Int64,2}:
-3 2 -5
-1 0 -2
3 -4 1
That seems to me to compute the transpose of B and not its adjoint. Instead, we should get this (from wikipedia):
and tried to get its adjoint using the adjoint() function which is mentionned in the Julia documentation here despite that the doc does not specifically said what this function does
julia> adjoint(B)
3×3 Adjoint{Int64,Array{Int64,2}}:
-3 -1 3
2 0 -4
-5 -2 1
Instead I want to get this:
In Matlab I get indeed:
>> adjoint(B)
ans =
-8.0000 18.0000 -4.0000
-5.0000 12.0000 -1.0000
4.0000 -6.0000 2.0000
Julia's adjoint is defined as the transpose of the complex conjugate of the input matrix. However, you seem to want the adjugate matrix:
The adjugate has sometimes been called the "adjoint", but today the "adjoint" of a matrix normally refers to its corresponding adjoint operator, which is its conjugate transpose.
You can compute the adjugate matrix by inverting, and then multiplying by the determinant:
julia> det(B) * inv(B)
3×3 Array{Float64,2}:
-8.0 18.0 -4.0
-5.0 12.0 -1.0
4.0 -6.0 2.0
Thanks to #Antoine Levitt and #Syx Pek on the Julia Slack for giving the suggestion of inverting and multiplying by determinant.
Original answer:
The adjugate matrix seems to be the transpose of the matrix of cofactors. Below is a naïve implementation of finding cofactors:
# import Pkg; Pkg.add("InvertedIndices")
using InvertedIndices # for cleaner code, you can remove this if you really want to.
function cofactor(A::AbstractMatrix, T = Float64)
ax = axes(A)
out = similar(A, T, ax)
for col in ax[1]
for row in ax[2]
out[col, row] = (-1)^(col + row) * det(A[Not(col), Not(row)])
end
end
return out
end
Then, to find the adjugate, you only need to transpose (transpose(cofactor(B))).
The answer is:
julia> cofactor(B, Float64) |> transpose
3×3 Transpose{Float64,Array{Float64,2}}:
-8.0 18.0 -4.0
-5.0 12.0 -1.0
4.0 -6.0 2.0
which is equivalent to what Matlab gives.
Edit: #Antoine Levitt on the Julia slack pointed out that this is essentially a rescaled inverse matrix, so if you figure out the scaling factor, you can just do inv(B) * scaling_factor (in the case of this matrix, it's 6).

R: What is wrong with rounding? [duplicate]

Yes I know why we always round to the nearest even number if we are in the exact middle (i.e. 2.5 becomes 2) of two numbers. But when I want to evaluate data for some people they don't want this behaviour. What is the simplest method to get this:
x <- seq(0.5,9.5,by=1)
round(x)
to be 1,2,3,...,10 and not 0,2,2,4,4,...,10.
Edit: To clearify: 1.4999 should be 1 after rounding. (I thought this would be obvious)
This is not my own function, and unfortunately, I can't find where I got it at the moment (originally found as an anonymous comment at the Statistically Significant blog), but it should help with what you need.
round2 = function(x, digits) {
posneg = sign(x)
z = abs(x)*10^digits
z = z + 0.5 + sqrt(.Machine$double.eps)
z = trunc(z)
z = z/10^digits
z*posneg
}
x is the object you want to round, and digits is the number of digits you are rounding to.
An Example
x = c(1.85, 1.54, 1.65, 1.85, 1.84)
round(x, 1)
# [1] 1.8 1.5 1.6 1.8 1.8
round2(x, 1)
# [1] 1.9 1.5 1.7 1.9 1.8
(Thanks #Gregor for the addition of + sqrt(.Machine$double.eps).)
If you want something that behaves exactly like round except for those xxx.5 values, try this:
x <- seq(0, 1, 0.1)
x
# [1] 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
floor(0.5 + x)
# [1] 0 0 0 0 0 1 1 1 1 1 1
As #CarlWitthoft said in the comments, this is the IEC 60559 standard as mentioned in ?round:
Note that for rounding off a 5, the IEC 60559 standard is expected to be used, ‘go to the even digit’. Therefore round(0.5) is 0 and round(-1.5) is -2. However, this is dependent on OS services and on representation error (since e.g. 0.15 is not represented exactly, the rounding rule applies to the represented number and not to the printed number, and so round(0.15, 1) could be either 0.1 or 0.2).
An additional explanation by Greg Snow:
The logic behind the round to even rule is that we are trying to
represent an underlying continuous value and if x comes from a truly
continuous distribution, then the probability that x==2.5 is 0 and the
2.5 was probably already rounded once from any values between 2.45 and 2.54999999999999..., if we use the round up on 0.5 rule that we learned in grade school, then the double rounding means that values
between 2.45 and 2.50 will all round to 3 (having been rounded first
to 2.5). This will tend to bias estimates upwards. To remove the
bias we need to either go back to before the rounding to 2.5 (which is
often impossible to impractical), or just round up half the time and
round down half the time (or better would be to round proportional to
how likely we are to see values below or above 2.5 rounded to 2.5, but
that will be close to 50/50 for most underlying distributions). The
stochastic approach would be to have the round function randomly
choose which way to round, but deterministic types are not
comforatable with that, so "round to even" was chosen (round to odd
should work about the same) as a consistent rule that rounds up and
down about 50/50.
If you are dealing with data where 2.5 is likely to represent an exact
value (money for example), then you may do better by multiplying all
values by 10 or 100 and working in integers, then converting back only
for the final printing. Note that 2.50000001 rounds to 3, so if you
keep more digits of accuracy until the final printing, then rounding
will go in the expected direction, or you can add 0.000000001 (or
other small number) to your values just before rounding, but that can
bias your estimates upwards.
This appears to work:
rnd <- function(x) trunc(x+sign(x)*0.5)
Ananda Mahto's response seems to do this and more - I am not sure what the extra code in his response is accounting for; or, in other words, I can't figure out how to break the rnd() function defined above.
Example:
seq(-2, 2, by=0.5)
# [1] -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
round(x)
# [1] -2 -2 -1 0 0 0 1 2 2
rnd(x)
# [1] -2 -2 -1 -1 0 1 1 2 2
Depending on how comfortable you are with jiggling your data, this works:
round(x+10*.Machine$double.eps)
# [1] 1 2 3 4 5 6 7 8 9 10
This method:
round2 = function(x, n) {
posneg = sign(x)
z = abs(x)*10^n
z = z + 0.5
z = trunc(z)
z = z/10^n
z*posneg
}
does not seem to work well when we have numbers with many digits. E.g. doing round2(2436.845, 2) will give us 2436.84. The issue seems to occur with the trunc(z) function.
Overall, I think it has something to do with the way R stores numbers and thus the trunc and float function doesn't always work. I was able to get around it in not the most elegant way:
round2 = function(x, n) {
posneg = sign(x)
z = abs(x)*10^n
z = z + 0.5
z = trunc(as.numeric(as.character(z)))
z = z/10^n
(z)*posneg
}
This mimics the rounding away from zero at .5:
round_2 <- function(x, digits = 0) {
x = x + abs(x) * sign(x) * .Machine$double.eps
round(x, digits = digits)
}
round_2(.5 + -2:4)
-2 -1 1 2 3 4 5

Multiply each component of vector by another vector (resulting in vector of length m*n)

Say I am making parts that come in three sizes, and each size has a certain tolerance:
target <- c(2, 4, 6)
tolerance <- c(0.95, 1.05)
What I'd like to end up with is an array that contains the limits of the tolerance for each target (i.e. target*0.95, target*1.05):
tol = (2*0.95, 2*1.05, 4*0.95, 4*1.05, 6*0.95, 6*1.05)
Here's a really ugly way of getting there, but I know there is a simple way to do this.
j<-1
tol<-NULL
for (i in target){
tol[j] <- i*tolerance[1]
tol[j+1] <- i*tolerance[2]
j<-j+2
}
The vector tol can be calculated using outer() like this:
tol <- c(outer(tolerance,target))
#> tol
#[1] 1.9 2.1 3.8 4.2 5.7 6.3
You can achieve that using matrix product:
target <- c(2, 4, 6)
tolerance <- c(0.95, 1.05)
target %*% t(tolerance)
[,1] [,2]
[1,] 1.9 2.1
[2,] 3.8 4.2
[3,] 5.7 6.3
The other answer would have my preference, but this alternative might generalise better in some specific context (more than two vectors)
Reduce("*", expand.grid(list(tolerance, target)))
Mostly for fun - using R's recycling:
rep(target, each = length(tolerance)) * tolerance
#[1] 1.9 2.1 3.8 4.2 5.7 6.3

Round to nearest arbitrary number from list

I'm basically looking for a way to do a variation of this Ruby script in R.
I have an arbitrary list of numbers (steps of a moderator for a regression plot in this case) which have unequal distances from each other, and I'd like to round values which are within a range around these numbers to the nearest number in the list.
The ranges don't overlap.
arbitrary.numbers <- c(4,10,15) / 10
numbers <- c(16:1 / 10, 0.39, 1.45)
range <- 0.1
Expected output:
numbers
## 1.6 1.5 1.4 1.3 1.2 1.1 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.39 1.45
round_to_nearest_neighbour_in_range(numbers,arbitrary.numbers,range)
## 1.5 1.5 1.5 1.3 1.2 1.0 1.0 1.0 0.8 0.7 0.6 0.4 0.4 0.4 0.2 0.1 0.4 1.5
I've got a little helper function that might do for my specific problem, but it's not very flexible and it contains a loop. I can post it here, but I think a real solution would look completely different.
The different answers timed for speed (on a million numbers)
> numbers = rep(numbers,length.out = 1000000)
> system.time({ mvg.round(numbers,arbitrary.numbers,range) })[3]
elapsed
0.067
> system.time({ rinker.loop.round(numbers,arbitrary.numbers,range) })[3]
elapsed
0.289
> system.time({ rinker.round(numbers,arbitrary.numbers,range) })[3]
elapsed
1.403
> system.time({ nograpes.round(numbers,arbitrary.numbers,range) })[3]
elapsed
1.971
> system.time({ january.round(numbers,arbitrary.numbers,range) })[3]
elapsed
16.12
> system.time({ shariff.round(numbers,arbitrary.numbers,range) })[3]
elapsed
15.833
> system.time({ mplourde.round(numbers,arbitrary.numbers,range) })[3]
elapsed
9.613
> system.time({ kohske.round(numbers,arbitrary.numbers,range) })[3]
elapsed
26.274
MvG's function is the fastest, about 5 times faster than Tyler Rinker's second function.
A vectorized solution, without any apply family functions or loops:
The key is findInterval, which finds the "space" in arbitrary.numbers where each element in numbers is "between". So, findInterval(6,c(2,4,7,8)) returns 2, because 6 is between the 2nd and 3rd index of c(2,4,7,8).
# arbitrary.numbers is assumed to be sorted.
# find the index of the number just below each number, and just above.
# So for 6 in c(2,4,7,8) we would find 2 and 3.
low<-findInterval(numbers,arbitrary.numbers) # find index of number just below
high<-low+1 # find the corresponding index just above.
# Find the actual absolute difference between the arbitrary number above and below.
# So for 6 in c(2,4,7,8) we would find 2 and 1.
# (The absolute differences to 4 and 7).
low.diff<-numbers-arbitrary.numbers[ifelse(low==0,NA,low)]
high.diff<-arbitrary.numbers[ifelse(high==0,NA,high)]-numbers
# Find the minimum difference.
# In the example we would find that 6 is closest to 7,
# because the difference is 1.
mins<-pmin(low.diff,high.diff,na.rm=T)
# For each number, pick the arbitrary number with the minimum difference.
# So for 6 pick out 7.
pick<-ifelse(!is.na(low.diff) & mins==low.diff,low,high)
# Compare the actual minimum difference to the range.
ifelse(mins<=range+.Machine$double.eps,arbitrary.numbers[pick],numbers)
# [1] 1.5 1.5 1.5 1.3 1.2 1.0 1.0 1.0 0.8 0.7 0.6 0.4 0.4 0.4 0.2 0.1 0.4 1.5
Yet another solution using findInterval:
arbitrary.numbers<-sort(arbitrary.numbers) # need them sorted
range <- range*1.000001 # avoid rounding issues
nearest <- findInterval(numbers, arbitrary.numbers - range) # index of nearest
nearest <- c(-Inf, arbitrary.numbers)[nearest + 1] # value of nearest
diff <- numbers - nearest # compute errors
snap <- diff <= range # only snap near numbers
numbers[snap] <- nearest[snap] # snap values to nearest
print(numbers)
The nearest in the above code is not really mathematically the nearest number. Instead, it is the largest arbitrary number such that nearest[i] - range <= numbers[i], or equivalently nearest[i] <= numbers[i] + range. So in one go we find the largest arbitrary number which is either in the snapping range for a given input number, or still too small for that. For this reason, we only need to check one way for snap. No absolute value required, and even the squaring from a previous revision of this post was unneccessary.
Thanks to Interval search on a data frame for the pointer at findInterval, as I found it there before recognizing it in the answer by nograpes.
If, in contrast to your original question, you had overlapping ranges, you could write things like this:
arbitrary.numbers<-sort(arbitrary.numbers) # need them sorted
range <- range*1.000001 # avoid rounding issues
nearest <- findInterval(numbers, arbitrary.numbers) + 1 # index of interval
hi <- c(arbitrary.numbers, Inf)[nearest] # next larger
nearest <- c(-Inf, arbitrary.numbers)[nearest] # next smaller
takehi <- (hi - numbers) < (numbers - nearest) # larger better than smaller
nearest[takehi] <- hi[takehi] # now nearest is really nearest
snap <- abs(nearest - numbers) <= range # only snap near numbers
numbers[snap] <- nearest[snap] # snap values to nearest
print(numbers)
In this code, nearestreally ends up being the nearest number. This is achieved by considering both endpoints of every interval. In spirit, this is very much like the version by nograpes, but it avoids using ifelse and NA, which should benefit performance as it reduces the number of branching instructions.
Is this what you want?
> idx <- abs(outer(arbitrary.numbers, numbers, `-`)) <= (range+.Machine$double.eps)
> rounded <- arbitrary.numbers[apply(rbind(idx, colSums(idx) == 0), 2, which)]
> ifelse(is.na(rounded), numbers, rounded)
[1] 1.5 1.5 1.5 1.3 1.2 1.0 1.0 1.0 0.8 0.7 0.6 0.4 0.4 0.4 0.2 0.1 0.4 1.5
Please note that due to rounding errors (most likely), I use range = 0.1000001 to achieve the expected effect.
range <- range + 0.0000001
blah <- rbind( numbers, sapply( numbers, function( x ) abs( x - arbitrary.numbers ) ) )
ff <- function( y ) { if( min( y[-1] ) <= range + 0.000001 ) arbitrary.numbers[ which.min( y[ -1 ] ) ] else y[1] }
apply( blah, 2, ff )
This is still shorter:
sapply(numbers, function(x) ifelse(min(abs(arbitrary.numbers - x)) >
range + .Machine$double.eps, x, arbitrary.numbers[which.min
(abs(arbitrary.numbers - x))] ))
Thanks #MvG
Another option:
arb.round <- function(numbers, arbitrary.numbers, range) {
arrnd <- function(x, ns, r){
ifelse(abs(x - ns) <= range +.00000001, ns, x)
}
lapply(1:length(arbitrary.numbers), function(i){
numbers <<- arrnd(numbers, arbitrary.numbers[i], range)
}
)
numbers
}
arb.round(numbers, arbitrary.numbers, range)
Yields:
> arb.round(numbers, arbitrary.numbers, range)
[1] 1.5 1.5 1.5 1.3 1.2 1.0 1.0 1.0 0.8 0.7 0.6 0.4 0.4 0.4 0.2 0.1 0.4 1.5
EDIT: I removed the return call at the end of the function as it's not necessary adn can burn time.
EDIT: I think a loop will be even faster here:
loop.round <- function(numbers, arbitrary.numbers, range) {
arrnd <- function(x, ns, r){
ifelse(abs(x - ns) <= range +.00000001, ns, x)
}
for(i in seq_along(arbitrary.numbers)){
numbers <- arrnd(numbers, arbitrary.numbers[i], range)
}
numbers
}

Resources