Can we sum a series of values,even some of them are "Inf"?If i want to do it,how can i do?

I get a series answer of optimal problem in several times,and i want to sum of them and average them,however, some of them are "Inf",can i or how to write a code to ignore those "Inf" and sum the others which is not "Inf" ?

Also,if i can sum them when some of them are "Inf",and don't know the amount of value which is "Inf",do we have some method to average them,i mean,ignore the "Inf" and average the others?

The version of matlab is 2015a

Answer 1

[John D'Errico]

No. You cannot add numbers if some of them are infs, at least not without getting an inf as a result. OR you might even get a NaN, since inf+ -inf will result in a NaN.

sum([1:5,inf,6,7,-inf])
ans =
   NaN

but

sum([1:5,inf,6,7,inf])
ans =
   Inf

You also cannot know that some infs are bigger than other infs, as I think you may have asked.

If you want to form a sum of numbers, excluding infs, this is trivial:

V = [1:5,inf,6,7,-inf];
sum(V(~isinf(V)))
ans =
    28

Also, if you have a reasonably recent MATLAB release, you will find that tools like mean allow you to ignore NaNs in the data as an option. So you could simply convert all infs into NaNs, then use mean with the correct option set. (Or use nanmean in older releases.)

[yang-En Hsiao]

yes!that what i want to ask,but now if i want to average the result in total loop,i mean ,you V=[1,2,3,4,5,inf,6,7,-inf].If the result is 1 in the first loop ,the result is 2 in the second loop,the result is inf and -inf in the sixth and ninth loop,if i use for to make this loop,and want to average them,how do i write the code?i mean the answer should be 28/7=4

[Torsten]

mean(V(~isinf(V)))

[yang-En Hsiao]

but i mean if situation is like this

V=

1

V=

2

V=

3

...

V=

-inf

not V=[1:5,inf,6,7,-inf]

If i write mean(V(~isinf(V))),it should be the same as sum(V(~isinf(V))),because it should be just number in every time (loop),so i think the answer will be like 1/1,2/1,3/1,etc

[John D'Errico]

You might think the answer should be that. But you would be wrong.

V = [1:5,inf,6,7,-inf];
mean(V(~isinf(V)))
ans =
     4

But perhaps I am misunderstanding your problem with the answer as given.

Are you asking as if V is a scalar, computed at each iteration inside a loop, perhaps by some complicated expressions? So then the answer is as it always should be, to just write the code properly. For example:

V=[1:5,inf,6,7,-inf];
S = 0;
nfinite = 0;
for n = 1:numel(V)
  % assume Vi is computed in some less boring way
  Vn = V(n);
  if ~isinf(Vn)
    S = S + Vn;
    nfinite = nfinite + 1;
  end
end
Mu = S/nfinite;

Mu
Mu =
     4

I don't see a problem. The code as I have suggested it should be is more efficient than a loop, but still entirely valid either way.

[yang-En Hsiao]

like this ,we

c=0
for k=1:10
     ******
     %optimal problem calculating code(code omit),it will produce optimal problem result
     *******
c=c+sum(~isinf(optimal problem result))%c is the summation of all optimal problem result
end

If i want to average the c,how do i write the code?if i just write as c_mean=mean(~isinf(optimal problem result)),you may find that c value will be equal to c_mean

if when k=1, optimal problem result=1

when k=2, optimal problem result=2

....

when k=5, optimal problem result=5

when k=6 ,optimal problem result=Inf

.....

when k=9 ,optimal problem result=-Inf

I want to average those optimal problem result

[Torsten]

c = zeros(10,1);
for k = 1:10
     **
     %optimal problem calculating code(code omit),it will produce optimal problem result
     *******
  c(k) = optimal problem result; 
end
c_mean = mean(c(~isinf(c)))

[John D'Errico]

I showed you exactly how to solve the problem. Or you can save the results for each iteration as Torsten did, then form the mean of the vector of results at the end. Both approaches work with no problems. You can do the sum in the loop or outside of the loop. If you do it inside the loop, just divide by the number of elements that were found to be finite at the end. What is the problem?

[yang-En Hsiao]

Here is my code,and i found a bug in here if i use the mean(c(~isinf(c)))

bd=2
for l=1:bd
%hat_p_up=P_p;
 hat_p_up=0.0824
%OP4
%declare
K=4;
N=4;
L=5;%distance between RX & TX
xi=10^-4%tolerence between 
nois_var_hk_2pow=0.1*(L^(-2.5))*10;%W,0.1*(L^(-2.5)),if this unit is dbm
w_nois_var_hk_2pow=10^(-3)*10^(0.1*nois_var_hk_2pow)
nois_var_ak_2pow=[1.0000e-10 1.0000e-10 1.0000e-10 1.0000e-10 ];
nois_var_dk_2pow=[1.0000e-08 1.0000e-08 1.0000e-08 1.0000e-08 ];
bar_r=[10 10 10 10]
P_T=10
co = zeros(2,1);
h_1=sqrt(w_nois_var_hk_2pow/2)*(randn(N,1)+1i*randn(N,1));
h_2=sqrt(w_nois_var_hk_2pow/2)*(randn(N,1)+1i*randn(N,1));
h_3=sqrt(w_nois_var_hk_2pow/2)*(randn(N,1)+1i*randn(N,1));
h_4=sqrt(w_nois_var_hk_2pow/2)*(randn(N,1)+1i*randn(N,1));
h_kk=cat(2,h_1 ,h_2 ,h_3, h_4)
 for n=1:4
    h_k{n}=h_kk(1:4 , n);
    n=n+1;
 end

%==========================
cvx_begin

    variable FNNK_up(N,N,K) semidefinite;%c7
    variable rho_k_up(1,1,K) semidefinite;
%==========================
Fkk_up=cat(2,FNNK_up);
    
    up=0
   for o_up=1:4
       Fk_up{o_up}=Fkk_up(1:4,o_up+3*up:4*o_up)
       up=up+1;
   end
   tr_ace_up=0
   for t=1:K
       tr_ace_up=tr_ace_up+trace(Fk_up{t})  
   end
%====================================    
%object function
    minimize( tr_ace_up )
%==================================== 
%Constraint 
subject to 
%c3
rho_k_up<=1;

%===================================================
%c5
       c5_left_hand_up = 0;
        for k = 1:K
            sum_5_up = 0;
            for j = 1:K
                if j ~= k  
                 sum_5_up = sum_5_up +  h_k{k}' * Fk_up{j} * h_k{k};
                end
            end      
            c5_left_hand_up = c5_left_hand_up - sum_5_up+ (h_k{k}' * Fk_up{k} * h_k{k}*inv_pos(bar_r(1)))
            c5_right_hand_up= nois_var_ak_2pow(1)+ ( nois_var_dk_2pow(1)*inv_pos(rho_k_up(k)) )
            real( c5_left_hand_up ) >=   c5_right_hand_up
        end

%===================================================   
%c10
    c10_left_hand_up = 0;           
            sum_10_up = 0;
            for j = 1:K                  
                 sum_10_up= sum_10_up +  h_k{k}' * Fk_up{j} * h_k{k}; 
            end
            c10_left_hand_up = c10_left_hand_up + sum_10_up+nois_var_ak_2pow(1)
            c10_right_hand_up=hat_p_up*inv_pos(1-rho_k_up(k))  
            real(c10_left_hand_up)>=  c10_right_hand_up    
cvx_end  
end
co(bd) = cvx_optval
c_sum=sum(co(~isinf(co)))
c_mean = mean(co(~isinf(co)))

I got two cvx_optval,one is Inf and one is 538.6262,so theoretically,the c_sum should be the same as c_mean,bacause i have only one value which is not inf,but i found that c_sum=538.6262,and c_mean= 269.3131,why is that?

By the way,if you want to run my code ,you have to install the cvx into the matlab first

[Torsten]

co is a vector with elements 0 and 538.6262. Taking the mean of this vector gives (0+538.6262)/2 = 269.3131, taking the sum of this vector gives 0+538.6262 = 538.6262.