Cumsum Function Reproduction

I wish to write a function that performs the same thing as the built in cumsum function and I can't figure out how to do it, does anyone have any hints or ideas about how I can write the code? Or how the cumsum function itself is writtten in the Matlab program?

Answer 1

[Jonathan]

[DELETED]

[Walter Roberson]

You've just given away a complete solution to homework :-(

[Jan]

@Jonathan: Solving homework questions is not helpful. Neither for the teachers, nor for the pupils, nor for the forum.

How can Anthony now submit a solution without telling the teacher, that it is a result of copy&paste?

[Jonathan]

Hmm, you are right. I normally catch that. I didn't consider that this time. Thanks for pointing that out guys.

I am well aware of the reasons to not post solutions to homework, but thanks for the lecture anyway. c-;

[Jan]

@Jonathan: Thanks for hidign the solution.

I've mentioned the "copy&paste" term mainly for Anthony.

Answer 2

[Jan]

CUMSUM is written in C and compiled. To create an equivalent function in Matlab you need a FOR loop. At first you start with the first element (or vector, if you want to operate on matrices). Then you add this cummulatively to the following elements / vectors.

I assume, this is a homework. You need three lines only, therefore it is hard to give further advices.

Answer 3

[Anthony]

For the person concerned with copy and pasting, I was asking for assistancem not answers, thank you for questioning my academic integrity, thank you for the help.

[Jan]

You have asked for "hints or ideas". Therefore your question welcome in this forum.

Only for the one, who want to answer, it is not trivial to estimate, how much help is still helpful for homework questions.

[Walter Roberson]

Anthony, the principles of the solution are so basic that once you had seen the solution, you would have had to go through a lot of trouble to deliberately invent a "perverse" solution just to avoid accusations of copying. You would have had to shown Jonathan's solution in your documentation, and then spent a bunch of time in your documentation proving that your own solution wasn't "really" copying his, which would have been a very difficult argument to make, as there just aren't many ways to program this functionality.

Answer 4

[Anthony]

Once again, wasn't asking for the solution. I'm trying to figure it out and it's not working. There aren't many places or people for me to turn to for assistance so I figured this would be a good place to start.

[Anthony]

And besides, I didn't even see Jonathan's answer, so no copying and pasting will be done here.

[Jan]

Everything is fine, Anthony. You show a lot of interested and without doubt yu are not a lazy cheater.

Let's come back to Matlab. Does my hint help, that you need a FOR loop? You can post, what you have tried so far and ask specific questions.

[Walter Roberson]

Right: Jan is indicating that you are now better protected because you can no longer be thought to have copy and pasted your solution.

Jan also posted an outline of what has to be done. There is not much more that can be added to what he said without effectively showing the code.

If you were to post what you have already, then people could look and perhaps indicate the concept that you missed.

Answer 5

[Anthony]

Okay well I've tried a couple different methods but here is what I've been working on now.

function a = mycumsum(x)
for i = 2:length(x)
  a    = zeros(1,length(x));
  a(1) = x(1);
  a(i) = x(i)+ x(i-1);
end

Output: a=[1 2 3 4]; mycumsum(a)= [1 0 0 7]

[Anthony]

Note: For a length 2 vector this actually works perfectly.

[Walter Roberson]

Do not reset the vector to zero each time through the loop ;-)

[Anthony]

oh ok! I'm getting closer now for the same vector [1 2 3 4] it's giving me [1 3 5 7] it looks like it's adding up each element before it so it's almmost there. I know i can make add more elements simply by adding x(i) + x(i-1) + x(i-2)...etc but I don't know how to generalize that statement so that I can use a vector with any number of elements, that's where I'm getting stuck.

[bym]

take a look at using the colon (:) operator!

[Walter Roberson]

The colon operator won't help much.

In the body of the loop, you are adding two values from the data vector to produce the result. Where are you accumulating the sum from the start of the vector?

[bym]

I was merely suggesting that the OP consider replacing the

a(i) = x(i)+ x(i-1);

portion of the loop with something to get the cumulative sum rather than the sum of the most recent 2 entries

[Jan]

@Anthony: Fine, the problem is solved partially!

Let's look, what happens inside the loop: In each iteration you create a new vector "a" overwriting the formerly calculated values. This is not wanted.

What about re-using "x" itself?

[Jonathan]

It looks like it would be good to write out what happens in each iteration of the for loop. Suppose you use the code above, modified by moving the "reset the vector to zero" as suggested below, like this.

function a = mycumsum(x)

a = zeros(1,length(x));

for i = 2:length(x)

a(1) = x(1);

a(i) = x(i)+ x(i-1);

end

For your input a = [1 2 3 4], the execution of mycumsum(a) looks (conceptually) like this inside the function.

x = [1 2 3 4]

a = zeros(1,length(x))

a(1) = x(1)

a(2) = x(2) + x(2-1)

a(1) = x(1)

a(3) = x(3) + x(3-1)

a(1) = x(1)

a(4) = x(4) + x(4-1)

How would you rewrite this (without a loop) to get the answer you want? Then we can help you make a loop out of it.

Answer 6

[Daniel Shub]

I can think of 3 independent ways to implement a basic cumsum (ignoring its ability to handle n-D arrays).

There is a way using a for loop that Jan suggested and Anthony is trying to use.

There is a way using recursion.

There is a way using matrix multiplication (hint tril).

There might also be ways to use accumarray, bsxfun, and arrayfun, although I cannot easily think of how to do it. I am not sure which way is most efficient.

[Jan]

Most efficient: multi-threaded C-mex with using the Intel Compiler due to the 80 bit storage of the accumulator. Using MSVC leads earlier to the well known rounding effects with [1e17, 1, -1e17].