I'm comparing ILMath.Vec() with MatLab's and I'm seeing significant rounding errors.
For example, if I create a vector (using Start:0, Step:1.2635048525911006, End:20700) for each system:
MatLab: [Start:Step:End]
ILNumerics: Vec<double>(Start, Step, End)
And then take the average abs difference, I get an average error of 1.56019608343883E-09. However, I if create the vector by hand (using multiplication), I get an average error of only 3.10973469197506E-13, 4 magnitudes smaller error.
After looking at ILNumerics' vec function (using Reflector), I think I know why the average error value is so large. The ILMath.vec() function is using addition vs. multiplication. Summing the step value 16,384 times is not the same thing as multiplying the step value x N (where N is the current loop count) 16,384 times! The addition's rounding errors add up very quickly!
Please consider fixing this issue.
Related
Why do I get different results when using the same code running in different version of MATLAB (2016 vs 2021) for sum(b.*x1) where b is single and x1 is double. How to avoid such error between MATLAB version?
MATLAB v.2021:
sum(b.*x1)
ans =
single
-0.0013286
MATLAB 2016
sum(b.*x1)
ans =
single
-0.0013283
In R2017b, they changed the behavior of sum for single-precision floats, and in R2020b they made the same changes for other data types too.
The change speeds up the computation, and improves accuracy by reducing the rounding errors. Simply put, previously the algorithm would just run through the array in sequence, adding up the values. The new behavior computes the sum over smaller portions of the array, and then adds up those results. This is more precise because the running total can become a very large number, and adding smaller numbers to it causes more rounding in those smaller numbers. The speed improvement comes from loop unrolling: the loop now steps over, say, 8 values at the time, and in the loop body, 8 running totals are computed (they don’t specify the number they use, the 8 here is an example).
Thus, your newer result is a better approximation to the sum of your array than the old one.
For more details (a better explanation of the new algorithm and the reason for the change), see this blog post.
Regarding how to avoid the difference: you could implement your own sum function, and use that instead of the builtin one. I would suggest writing it as a MEX-file for efficiency. However, do make sure you match the newer behavior of the builtin sum, as that is the better approximation.
Here is an example of the problem. Let's create an array with N+1 elements, where the first one has a value of N and the rest have a value of 1.
N = 1e8;
a = ones(N+1,1,'single');
a(1) = N;
The sum over this array is expected to be 2*N. If we set N large enough w.r.t. the data type, I see this in R2017a (before the change):
>> sum(a)
ans =
single
150331648
And I see this in R2018b (after the change for single-precision sum):
>> sum(a)
ans =
single
199998976
Both implementations make rounding errors here, but one is obviously much, much closer to the expected result (2e8, or 200000000).
Maple helpfully can work out the solution to Laplace's equation in a square region and give me the answer in closed form (in terms of an infinite sum). If I try to plot the function of two variables as a 3d plot it gives me most of the surface but not all of it:
Here is the Maple code which produces the solution and turns it into an expression suitable for plotting
lapeq:=diff(v(x,y),x$2)+diff(v(x,y),y$2)=0;
bcs:=v(x,0)=0,v(0,y)=0,v(1,y)=0,v(x,1)=100;
sol1:=pdsolve({lapeq,bcs});
vxy:=eval(v(x,y),sol1);
the result of which is
All good so far. Plotting it via
plot3d(vxy,x=0..1,y=0..1);
gives a result which is fine for x in the full range (0<x<1) but only for y between 0 and around 0.9:
I have tried to evalf some point in the unknown region and Maple can't tell me numerical values there. Is there any way to get Maple to "try a bit harder" to evaluate those numbers?
You could try setting the number of terms in the sum
Compare
lapeq:=diff(v(x,y),x$2)+diff(v(x,y),y$2)=0;
bcs:=v(x,0)=0,v(0,y)=0,v(1,y)=0,v(x,1)=100;
sol1:=pdsolve({lapeq,bcs});
vxy:=subs(infinity=100,sol1);
plot3d(rhs(vxy),x=0..1,y=0..1);
With
restart;
lapeq:=diff(v(x,y),x$2)+diff(v(x,y),y$2)=0;
bcs:=v(x,0)=0,v(0,y)=0,v(1,y)=0,v(x,1)=100;
sol1:=pdsolve({lapeq,bcs});
vxy:=eval(v(x,y),sol1);
plot3d(vxy,x=0..1,y=0..1);
I'm not a huge fan of chopping the infinite sum at some value of the upper bound for n, without at least demonstrating either symbolically or numerically that it is justified. Ie, that chopping does not provide a false idea of convergence.
So, you asked how to make it work "harder". I'll take that to mean that you too might prefer to let evalf/Sum itself decide whether each infinite numeric sum converges -- rather than manually truncate it yourself at some finite value for the upper value of the range for n.
For fun, and caution, I also divide both numerator and denominator of K by the potentially large exp call (potentially much larger than 1). That may not be necessary here.
restart;
lapeq:=diff(v(x,y),x$2)+diff(v(x,y),y$2)=0:
bcs:=v(x,0)=0,v(0,y)=0,v(1,y)=0,v(x,1)=100:
sol1:=pdsolve({lapeq,bcs}):
vxy:=eval(v(x,y),sol1):
K:=op(1,vxy):
J:=simplify(combine(numer(K)/exp(2*Pi*n)))
/simplify(combine(denom(K)/exp(2*Pi*n))):
F:=subs(__d=J,
proc(x,y) local k, m, n, r;
if y<0.8 then
r:=Sum(__d,n=1..infinity);
else
UseHardwareFloats:=false;
m := ceil(1*abs(y/0.80)^16);
r:=add(Sum(eval(__d,n=m*n-k),n=1..infinity),
k=0..m-1);
end if;
evalf(r);
end proc):
plot3d( F, 0..1, 0..0.99 );
Naturally this is slower than mere chopping of terms to obtain a finite sum. And you might be satisfied with some technique that establishes that the excluded terms' sums are negligible.
I am working on a code in Least Square Non Negative solution recovery context on Matlab, and I need (with no more details because it's not that important for this question) to know the number of non zero elements in my matrices and arrays.
The function NNZ on matlab does exactly what I want, but it happens that I need more information about what Matlab thinks of a "zero element", it could be 0 itself, or the numerical zero like 1e-16 or less.
Does anybody has this information about the NNZ function, cause I couldn't get the original script
Thanks.
PS : I am not an expert on Matlab, so accept my apologies if it's a really simple task.
I tried "open nnz", on Matlab but I only get a small script of commented code lines...
Since nnz counts everything that isn't an exact zero (i.e. 1e-100 is non-zero), you just have to apply a relational operator to your data first to find how many values exceed some tolerance around zero. For a matrix A:
n = nnz(abs(A) > 1e-16);
Also, this discussion of floating-point comparison might be of interest to you.
You can add in a tolerance by doing something like:
nnz(abs(myarray)>tol);
This will create a binary array that is 1 when abs(myarray)>tol and 0 otherwise and then count the number of non-zero entries.
Having read carefully the previous question
Random numbers that add to 100: Matlab
I am struggling to solve a similar but slightly more complex problem.
I would like to create an array of n elements that sums to 1, however I want an added constraint that the minimum increment (or if you like number of significant figures) for each element is fixed.
For example if I want 10 numbers that sum to 1 without any constraint the following works perfectly:
num_stocks=10;
num_simulations=100000;
temp = [zeros(num_simulations,1),sort(rand(num_simulations,num_stocks-1),2),ones(num_simulations,1)];
weights = diff(temp,[],2);
I foolishly thought that by scaling this I could add the constraint as follows
num_stocks=10;
min_increment=0.001;
num_simulations=100000;
scaling=1/min_increment;
temp2 = [zeros(num_simulations,1),sort(round(rand(num_simulations,num_stocks-1)*scaling)/scaling,2),ones(num_simulations,1)];
weights2 = diff(temp2,[],2);
However though this works for small values of n & small values of increment, if for example n=1,000 & the increment is 0.1% then over a large number of trials the first and last numbers have a mean which is consistently below 0.1%.
I am sure there is a logical explanation/solution to this but I have been tearing my hair out to try & find it & wondered anybody would be so kind as to point me in the right direction. To put the problem into context create random stock portfolios (hence the sum to 1).
Thanks in advance
Thank you for the responses so far, just to clarify (as I think my initial question was perhaps badly phrased), it is the weights that have a fixed increment of 0.1% so 0%, 0.1%, 0.2% etc.
I did try using integers initially
num_stocks=1000;
min_increment=0.001;
num_simulations=100000;
scaling=1/min_increment;
temp = [zeros(num_simulations,1),sort(randi([0 scaling],num_simulations,num_stocks-1),2),ones(num_simulations,1)*scaling];
weights = (diff(temp,[],2)/scaling);
test=mean(weights);
but this was worse, the mean for the 1st & last weights is well below 0.1%.....
Edit to reflect excellent answer by Floris & clarify
The original code I was using to solve this problem (before finding this forum) was
function x = monkey_weights_original(simulations,stocks)
stockmatrix=1:stocks;
base_weight=1/stocks;
r=randi(stocks,stocks,simulations);
x=histc(r,stockmatrix)*base_weight;
end
This runs very fast, which was important considering I want to run a total of 10,000,000 simulations, 10,000 simulations on 1,000 stocks takes just over 2 seconds with a single core & I am running the whole code on an 8 core machine using the parallel toolbox.
It also gives exactly the distribution I was looking for in terms of means, and I think that it is just as likely to get a portfolio that is 100% in 1 stock as it is to geta portfolio that is 0.1% in every stock (though I'm happy to be corrected).
My issue issue is that although it works for 1,000 stocks & an increment of 0.1% and I guess it works for 100 stocks & an increment of 1%, as the number of stocks decreases then each pick becomes a very large percentage (in the extreme with 2 stocks you will always get a 50/50 portfolio).
In effect I think this solution is like the binomial solution Floris suggests (but more limited)
However my question has arrisen because I would like to make my approach more flexible & have the possibility of say 3 stocks & an increment of 1% which my current code will not handle correctly, hence how I stumbled accross the original question on stackoverflow
Floris's recursive approach will get to the right answer, but the speed will be a major issue considering the scale of the problem.
An example of the original research is here
http://www.huffingtonpost.com/2013/04/05/monkeys-stocks-study_n_3021285.html
I am currently working on extending it with more flexibility on portfolio weights & numbers of stock in the index, but it appears my programming & probability theory ability are a limiting factor.......
One problem I can see is that your formula allows for numbers to be zero - when the rounding operation results in two consecutive numbers to be the same after sorting. Not sure if you consider that a problem - but I suggest you think about it (it would mean your model portfolio has fewer than N stocks in it since the contribution of one of the stocks would be zero).
The other thing to note is that the probability of getting the extreme values in your distribution is half of what you want them to be: If you have uniformly distributed numbers from 0 to 1000, and you round them, the numbers that round to 0 were in the interval [0 0.5>; the ones that round to 1 came from [0.5 1.5> - twice as big. The last number (rounding to 1000) is again from a smaller interval: [999.5 1000]. Thus you will not get the first and last number as often as you think. If instead of round you use floor I think you will get the answer you expect.
EDIT
I thought about this some more, and came up with a slow but (I think) accurate method for doing this. The basic idea is this:
Think in terms of integers; rather than dividing the interval 0 - 1 in steps of 0.001, divide the interval 0 - 1000 in integer steps
If we try to divide N into m intervals, the mean size of a step should be N / m; but being integer, we would expect the intervals to be binomially distributed
This suggests an algorithm in which we choose the first interval as a binomially distributed variate with mean (N/m) - call the first value v1; then divide the remaining interval N - v1 into m-1 steps; we can do so recursively.
The following code implements this:
% random integers adding up to a definite sum
function r = randomInt(n, limit)
% returns an array of n random integers
% whose sum is limit
% calls itself recursively; slow but accurate
if n>1
v = binomialRandom(limit, 1 / n);
r = [v randomInt(n-1, limit - v)];
else
r = limit;
end
function b = binomialRandom(N, p)
b = sum(rand(1,N)<p); % slow but direct
To get 10000 instances, you run this as follows:
tic
portfolio = zeros(10000, 10);
for ii = 1:10000
portfolio(ii,:) = randomInt(10, 1000);
end
toc
This ran in 3.8 seconds on a modest machine (single thread) - of course the method for obtaining a binomially distributed random variate is the thing slowing it down; there are statistical toolboxes with more efficient functions but I don't have one. If you increase the granularity (for example, by setting limit=10000) it will slow down more since you increase the number of random number samples that are generated; with limit = 10000 the above loop took 13.3 seconds to complete.
As a test, I found mean(portfolio)' and std(portfolio)' as follows (with limit=1000):
100.20 9.446
99.90 9.547
100.09 9.456
100.00 9.548
100.01 9.356
100.00 9.484
99.69 9.639
100.06 9.493
99.94 9.599
100.11 9.453
This looks like a pretty convincing "flat" distribution to me. We would expect the numbers to be binomially distributed with a mean of 100, and standard deviation of sqrt(p*(1-p)*n). In this case, p=0.1 so we expect s = 9.4868. The values I actually got were again quite close.
I realize that this is inefficient for large values of limit, and I made no attempt at efficiency. I find that clarity trumps speed when you develop something new. But for instance you could pre-compute the cumulative binomial distributions for p=1./(1:10), then do a random lookup; but if you are just going to do this once, for 100,000 instances, it will run in under a minute; unless you intend to do it many times, I wouldn't bother. But if anyone wants to improve this code I'd be happy to hear from them.
Eventually I have solved this problem!
I found a paper by 2 academics at John Hopkins University "Sampling Uniformly From The Unit Simplex"
http://www.cs.cmu.edu/~nasmith/papers/smith+tromble.tr04.pdf
In the paper they outline how naive algorthms don't work, in a way very similar to woodchips answer to the Random numbers that add to 100 question. They then go on to show that the method suggested by David Schwartz can also be slightly biased and propose a modified algorithm which appear to work.
If you want x numbers that sum to y
Sample uniformly x-1 random numbers from the range 1 to x+y-1 without replacement
Sort them
Add a zero at the beginning & x+y at the end
difference them & subtract 1 from each value
If you want to scale them as I do, then divide by y
It took me a while to realise why this works when the original approach didn't and it come down to the probability of getting a zero weight (as highlighted by Floris in his answer). To get a zero weight in the original version for all but the 1st or last weights your random numbers had to have 2 values the same but for the 1st & last ones then a random number of zero or the maximum number would result in a zero weight which is more likely.
In the revised algorithm, zero & the maximum number are not in the set of random choices & a zero weight occurs only if you select two consecutive numbers which is equally likely for every position.
I coded it up in Matlab as follows
function weights = unbiased_monkey_weights(num_simulations,num_stocks,min_increment)
scaling=1/min_increment;
sample=NaN(num_simulations,num_stocks-1);
for i=1:num_simulations
allcomb=randperm(scaling+num_stocks-1);
sample(i,:)=allcomb(1:num_stocks-1);
end
temp = [zeros(num_simulations,1),sort(sample,2),ones(num_simulations,1)*(scaling+num_stocks)];
weights = (diff(temp,[],2)-1)/scaling;
end
Obviously the loop is a bit clunky and as I'm using the 2009 version the randperm function only allows you to generate permutations of the whole set, however despite this I can run 10,000 simulations for 1,000 numbers in 5 seconds on my clunky laptop which is fast enough.
The mean weights are now correct & as a quick test I replicated woodchips generating 3 numbers that sum to 1 with the minimum increment being 0.01% & it also look right
Thank you all for your help and I hope this solution is useful to somebody else in the future
The simple answer is to use the schemes that work well with NO minimum increment, then transform the problem. As always, be careful. Some methods do NOT yield uniform sets of numbers.
Thus, suppose I want 11 numbers that sum to 100, with a constraint of a minimum increment of 5. I would first find 11 numbers that sum to 45, with no lower bound on the samples (other than zero.) I could use a tool from the file exchange for this. Simplest is to simply sample 10 numbers in the interval [0,45]. Sort them, then find the differences.
X = diff([0,sort(rand(1,10)),1]*45);
The vector X is a sample of numbers that sums to 45. But the vector Y sums to 100, with a minimum value of 5.
Y = X + 5;
Of course, this is trivially vectorized if you wish to find multiple sets of numbers with the given constraint.
as i know to get zero mean vector from given vector,we should substract mean of given vector from each memeber of this vector.for example let us see following example
r=rand(1,6)
we get
0.8687 0.0844 0.3998 0.2599 0.8001 0.4314
let us create another vector s by following operation
s=r-mean(r(:));
after this we get
0.3947 -0.3896 -0.0743 -0.2142 0.3260 -0.0426
if we calculate mean of s by following formula
mean(s)
we get
ans =
-5.5511e-017
actually as i have checked this number is very small
-5.5511*exp(-017)
ans =
-2.2981e-007
so we should think that our vector has mean zero?so it means that that small deviation from 0 is because of round off error?for exmaple when we are creating white noise or such kind off random uncorrelated sequence of data,actually it is already supposed that even for such small data far from 0,it has zero mean and it is supposed in this case that for example for this case
-5.5511e-017 =0 ?
approximately of course
e-017 means 10 to the power of -17 (10^-17) but still the number is very small and hypothetically it is 0. And if you type
format long;
you will see the real precision used by Matlab
Actually you can refer to the eps command. Although matlab uses double that can encode numbers down to 2.2251e-308 the precission is determined size of the number.
Use it in the format eps(number) - it tell you the how large is the influence of the least significant bit.
on my machine eg. eps(0.3) returns 5.5511e-17 - exactly the number you report.