NaN when calculating average

I read data from a .xls file using xlsread. There are some 3000 rows with price data, all in number format in excel. I further try to calculate the average of previous 10 days for every day using mean2(prices(i-10,1):prices(i,1)). I do this for last 2990 rows. surprisingly a lot of these throw NaN. I have also run a check "isnan" on each element in prices and none of them tests positive. also stumbled on something:

isnumeric(NaN)

ans =

     1
how is this working??

Answer 1

[David Young]

Although a NaN is not a number, it is of class double, and so it is numeric. If a NaN was not numeric, it could not be held in a matrix with other numbers.

Your problem is that prices(i-10,1):prices(i,1) may be empty, so mean2 returns NaN. In fact, that expression doesn't look likely to be correct. I suspect you intended to use mean2(prices(i-10:i, 1)).

Answer 2

[Andrew Newell]

You could use nanmean to ignore the NaN's while calculating means:

prices = rand(20,20);
n = size(prices,1); m = size(prices,2); window=10;
runningMean = zeros(n-window,m);
for ii=1:n-window
  runningMean(ii,:) = nanmean(runningMean(ii:ii+window-1,:));
end

Note that I am interpreting your statement of "previous 10 days" as the ten days before the current day. Your version is actually averaging 11 days.

[David Young]

Andrew: I don't think he or she has got NaNs in his or her raw data. The NaNs are being generated because he or she is taking the mean of an empty matrix, which happens (in his or her code) whenever the price has decreased in any 10-day stretch.

[Atakan]

Hi David;

I want to write a Matlab code for generating “m” numbers from N(0,1) using following algorithm. My code should be a general code for “m”, “mu” & “sigma”.

1.Generate u1, u2 from UNIF(0,1), then set y=tan(pi*(u1-1/2))

2.If u2 <= (sqrt(e)/2)*(1+y^2)*e^((-y^2)/2) then set x=y, otherwise go to step 1

3.Repeat 1-2 until you generate m numbers.

[David Young]

Atakan: It really does need to be in a different question, as it's not related to this one. Even if it was, I would expect you to show what you'd done so far, and explain where you get stuck.

[Atakan]

I am trying to ask it. But it gives an error such that "we are sorry but something went wrong". What can I do? Can you ask it for me? This is very important for me...

[Atakan]

This is my code:

function [randnormal]=atakan(a,b,m)

randnormal=[];

count=1;

while (count<=m)

R = normrnd(a,b);

u1=unifrnd(0,1);

u2=unifrnd(0,1);

y=tan(pi*(u1-1/2));

if (u2<=((sqrt(exp(1))/2)*(1+y^2)*(exp(1)^(-y^2/2))))

randnormal=[R;y];

end

count=count+1;

end

[Pankaj]

Thanks Andrew - good solution - it helps!

Answer 3

[Atakan]

Hi; I want to ask a question to the main page. But it gives an error such that "we are sorry but something went wrong". What can I do?

[Andrew Newell]

I tried creating a test question and had no problem. I would suggest trying again.

[Atakan]

I have tried many many times. But there is still a problem. If I send the question to you, can you ask the question for me? Because this is very important for me. thanks...

[Jan]

Dear Atakan: The servers have severe problems currently.