I have some data at hand, I want to correlate them, but I do not know what is the equation I should need, how can I do it ?

I have a lot data at hand, I want to correlate them, but I do not know what is the equation I should need, how can I do it ? Does anyone know how to deal with it ?

[dpb]

Plot 'em to start with...

[huazai2020]

em?

[dpb]

them (the data).

Those against what you think they should correlate with...

[Image Analyst]

No idea since you didn't explain your problem nearly well enough.

Try corr() or xcorr().

If you need more help (and I expect you will), then attach your data but ONLY after you read this link (please!)

[dpb]

NB: the above are linear correlation estimates only...w/o [without] knowing something about data structure you're shooting in the dark...

[huazai2020]

what is w/o？

[huazai2020]

Could you tell me what method should I use ?

[dpb]

No, because I can't see your data, either...nor do I have even whatever knowledge about what the data might happen to be that you have by being your data -- that should tell you something about what relationships might be.

Answer 1

[Puru Kathuria]

Hi,

Assuming your data has more than one features/variables, say M features and N data points.

Now, correlation can be observed between any pair of features and it can tell you how strongly are those variables related.

Correlation among those variables can be positive(when both variables grow in the same direction), negative(when variables grow in the opposite direction) or no correlation at all.

Now that we know something about correlation, there are multiple ways to observe it in the data.

First, you can plot your pair of variables and analyse the plots visually, this can be done using pair plotting [Link1, Link2, Link3]. Using this technique you will see C(M,2) number of plots, thus we prefer this plot when we have a limited number of variables to analyse.

Second, you can find a linear correlation coefficient, Pearson’s coefficient[Link4] is the most commonly used. This acts as a metric to calculate the correlation among variables. This correlation coefficient ranges from −1 to 1, where 1 implies perfect positive correlation, -1 implies a perfect negative correlation and 0 implies no correlation. [Link5]

Note: For MATLAB specific implementation check out the above mentioned links in the name of Link1, Link2, Link3, Link4 and Link5.

[huazai2020]

Could you show me the code for me ?

[dpb]

"Now, correlation can be observed between any pair of features and it can tell you how strongly are those variables related."

x=linspace(-1,1); y=x.^2;           % a set of fully-correlated variables

>> corrcoef(x,y)
ans =
    1.0000   -0.0000
   -0.0000    1.0000
>> 

Pearson's correlation isn't necessarily that informative all by itself...as noted before, it measures ONLY linear correlation.

What would you do with the above without plotting, given only x, y?

[Puru Kathuria]

data = readmatrix("data.xlsx"); %read data from the file


% plotmatrix Returns a plot where every pair of variables are plotted against each other. If the 
% number of variables are N then the plot will have N^2 sub plots where diagonals of the 
% plot shows the probability distribution of the variable
plotmatrix(data) 

% corrplot creates a matrix of plots showing correlations
%  among pairs of variables in input data.
corrplot(data)

[huazai2020]

It gives me errors for your code， my matlab is 2012b

data = readmatrix('data.xlsx'); %read data from the file

% plotmatrix Returns a plot where every pair of variables are plotted against each other. If the

% number of variables are N then the plot will have N^2 sub plots where diagonals of the

% plot shows the probability distribution of the variable

plotmatrix(data) ;

Undefined function 'readmatrix' for input arguments of type 'char'.

[huazai2020]

@dpb， the Value is dependent on the three values in data.xlsx,how do you know they are linear?

[huazai2020]

Have you received my message?