How to subtract the elements in one matrix with the corresponding elements in another matrix?

I'm doing a project work at the university in a course about Statistic methods of Earth Science. In this project we shall analyze a datafile with gravity measurements from the northern part of Sweden. The data file contains totally 3882 measure points. A part of the project work is to randomly exclude 300 measurements points from the original data file. A new matrix shall then be created where the 300 randomly selected measurement points have been deleted from the original data matrix. How to implement this in matlab? I tried this code:

clear

data = load('Mala_grav.txt');

Gravity = data(:,3); % data file with gravity measurements

[y,idx] = datasample(Gravity,300); % randomly pick out 300 measurement points

idx2 = sort(idx); % sort the index in increasing order

for i=1:length(Gravity)

for j=1:1:length(idx2)

if i==idx2(j) % compare index in the original data with the index in the vector with the 300 randomly chosen measurements points

G2(i)=0; % set value to zero if i = j

else

G2(i)=Gravity(i);

end

end

end

--------------

Problem: The 300 randomly chosen data points are not excluded in the new data matrix G2. How to subtract these values from the elements of the original data matrix?

Hope someone can give me a clue to solve this problem?!

Thanks in advance!

/Mattias

Answer 1

[Bjorn Gustavsson]

You're not asking for values to subtract, but indices to not include. My simple suggestion is that you take a look at the randperm function. You should be able to use that function to generate an array of indices to not include. You should be able to adapt this snippet:

data = randn(17,3);
idx2all = randperm(size(data,1));
idx2skip = idx2all(1:5);
data5shorter = data;
data5shorter(idx2skip,:) = [];

HTH

[Mattias Larsson]

Thanks for reply to my problem. I rewrote my script as follows:

clear

data = load('Mala_grav.txt');

Gravity = data(:,3);

orig = length(Gravity);

todelete = 300;

rndidx = randperm(orig,todelete);

--------------

Now I have a row vector with the index for 300 randomly selected values from the original data file (Gravity).

I understand how to randomly pick out 300 values from the original data file but the problem is then what to do to set the corresponding values to zero based on the index in the 'rndidx'-vector? So, I want to create a new vector ... where the elements for selected indexes are set to zero. The vector should be of the same length as the original data file (3882 posts). Our teacher want us to plot the data in the modified dataset.

[Bjorn Gustavsson]

OK, (I have opinions on your teachers "wants"), then your objective is the oposite to what I guessed, when you want to do partioning of your data it is typically better to select a sub-set and work with that. With the shorter data you can then do all statistical tests and estimates without to much thought, if you start to insert zeroes or nans you will have to start keeping track of those and everything becomes more cumbersome.

To insert zeros in the place of the 300 unwanted you can simply modify my snippet to:

data5shorter(idx2skip,:) = 0;

to set the elements to zero. For ploting it migth be prettier to set them to not-a-numbers:

data5shorter(idx2skip,:) = nan;

HTH

[Mattias Larsson]

Thanks for your advices.

I have had a follow-up conversation with my teacher about the project. It turned out that the 300 randomly selected measurement points should be deleted from the original data file. Sorry for my confusion!

The new matrix will therefore consist of 3882-300 = 3582 measurement points. The original data file consists of three columns. 1. X-position, 2. Y-position, 3. Gravity data (for the actual position).

What needs to be done then, is to subtract the 300 randomly selected measurement points from the original data set. What is the smartest way to do this?

[Bjorn Gustavsson]

Then you should simply modify the random index-selection in my original example to something like this:

idx2remove = randperm(3882,300); % 300 random unique integers between 1 and 3882
data(idx2remove,:) = [];         % remove those rows from your data-set.

But don't remove original data on disk for real. That is a precious resource that once did cost resources to obtain. 300 x N_cols worth of storage for data in long-double-format costs 0 öre. In your workspace do whatever. I'd prefer to generate a secondary data-array with 300-rows-less to work with.

[Mattias Larsson]

Regarding your comment: "I'd prefer to generate a secondary data-array with 300-rows-less to work with."

Do you mean something like this:

clear

data = load('Mala_grav.txt');

m = length(data); % number of rows in the original data file

b = zeros(m-300,3); % create a new matrix with 300-less-rows.

And what's the meaning of the brackets on the right side in the following?

data(idx2remove,:) = [];

[Bjorn Gustavsson]

Well I'd typically do it like this:

data = load('Mala_grav.txt');
working_data = data;
idx2remove = randperm(3882,300);
working_data(idx2remove,:) = [];
whos data working_data % just to check that everything works as planned

The meaning of the brackets is an empty array. The meaning of that entire line is that the entire idx2remove rows in the working_data array will be set to the empty array - i.e. they dissapear. Also, did you not see that when you tried my suggestion?

HTH

[Mattias Larsson]

Thanks, now it became much clearer to me. At first, I couldn't understand why it was an empty matrix on the right side. I ran the script in my MATLAB program. Received the following results:

Name Size Bytes Class Attributes

data 3882x3 93168 double

working_data 3582x3 85968 double

------

It works. Fantastic!

[Bjorn Gustavsson]

Good.

(Sound programming advice: Try to make small test-cases that you can run and look at the results element-by-element and simple cases that you can figure out manually. That way you have an easy way to check that your functions have the correct behaviour. Then you can scale up in size and whatnots. Apply as appropriate.)

[Mattias Larsson]

Ok. I will try to follow this strategy next I get trouble with my matlab programming. :-)