K-mean Clustering

Hi Everyone, can someone help me on how to use the K-mean clustering or perhaps share with me the suitable coding use to cluster wind speed data. I hava wind speed data in the form of Latitude, Longitude, Wind Speed. I want to cluster the data into 3 groups.

Answer 1

[Image Analyst]

If you have all the lat and lon values, then just put each into kmeans separately:

numColumns = 26; % Or however many columns you know there to be.
[xIndexes, xCentroids] = kmeans(lon, numColumns);
numRows = 50; % Or however many rows you know there to be.
[yIndexes, yCentroids] = kmeans(lat, numRows);

The values of the columns (x or longitude values) will be in xCentroids.

The values of the rows (y or lat values) will be in yCentroids.

[MAT NIZAM UTI]

So there will be different centroids based on the Lat Long ? or it will produce the same centroids ?

[Image Analyst]

Youi will have different locations. Imagine that you ran lines through every column and every row of your spots. Doing this will get you the x locations of every column, and the y location of every row. Isn't that what you want to achieve with kmeans?

[Image Analyst]

, it doesn't make sense. Why is Y1 random? And why is Y1 a row vector while X1 is a column vector? Even if Y1 was also a column vector, it doesn't make sense to cluster random data.

And where is K in your kmeans() call? You read in the badly-named "k" but don't even consider it when you're doing kmeans? Did you realize you're calling kmeans without your data???

I would have fixed it for you but I realized I don't know what each row of k represents.

% Demo by Image Analyst
clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
fontSize = 22;
% Read in data
k = readmatrix('WIND_26YEARS.csv');
% Plot raw data
subplot(3, 1, 1);
plot(k, 'b-')
grid on;
xlabel('index', 'FontSize',fontSize);
ylabel('Value of k', 'FontSize',fontSize)
title('All the k Values', 'FontSize',fontSize)
% Plot histogram of k data.
subplot(3, 1, 2);
histogram(k);
grid on;
xlabel('k', 'FontSize',fontSize);
ylabel('Count', 'FontSize',fontSize)
title('Distribution of k.  Note no clusters!', 'FontSize',fontSize)
% Original poster's (bad) code below:
subplot(3, 1, 3);
X1=(1:6943)';
Y1=randn(6943,1);
numClusters=3;
idx1=kmeans([X1, Y1],numClusters,'Replicates',5);
pointclust=repmat(idx1,1,numClusters)==repmat(1:numClusters,numel(idx1),1);
colors=hsv(numClusters);
for j=1:numClusters
    plot(X1(pointclust(:,j)),Y1(pointclust(:,j)),'Color',colors(j,:));
    if j==1
        hold on;
    end
end
hold off;
xlabel('X1', 'FontSize',fontSize);
ylabel('Y1', 'FontSize',fontSize)
title('Clusters are in different colors', 'FontSize',fontSize)
grid on;
g = gcf;
g.WindowState = 'maximized'

[MAT NIZAM UTI]

Okay, thanks a lot for your help .. I tried to modified which part that I want to use and also suitable for my data. I tried a few coding because this is also my first time use Matlab for CLustering.

[Image Analyst]

@MAT NIZAM UTI, I can continue to help but you'd have to tell me how to split apart your data. There might be clusters but until your data is organized correctly they may not be evident. Again, what does each row of your data represent? Can it be divided evenly into a number of subsets? Like morning and evening windspeeds, or by month or something?

[MAT NIZAM UTI]

Since my data is gridded, the actual format for my wind speed data is

Lat1, Long1, Wind Speed Value2

Lat2, Long2, Wind Speed Value2

and continue until last latitude and longitude

So the value of wind speed is actually the monthly average of wind speed on each latitude and longitude during the northeast monsoon. So one point of location (Latitude, Longitude) only consist with one value of wind speed as shown in the excel data. But I only provide the wind speed value.

So the idea of the clustering (K-mean) is to produce 3 clusters of data based on the wind speed value.

[Image Analyst]

Then why is the data not a multiple of 3?

k = readmatrix('WIND_26YEARS.csv');
lats = k(1:3:end); % 2315 long
lons = k(2:3:end); % 2314 long
speeds = k(3:3:end); % 2314 long

[MAT NIZAM UTI]

Do you mean 'multiple of 3' is refer to wind speed value times 3 (Wind speed x 3) or refer to the 3 column ?

[Image Analyst]

If the format is as you said, all arrays (lats, lons, and speeds) should be the same length, right? Why is lats one element longer?

[MAT NIZAM UTI]

Based on the data i has shared with you, both lattitude and longitude have the same number of column which is 3527 number of rows.

[Image Analyst]

I think you uploaded a different data file than you think. Look what happens when I run this code:

% Read in data
k = readmatrix('WIND_26YEARS.csv');
k = readmatrix('WIND_26YEARS.csv');
lats = k(1:3:end); % 2315 long
lons = k(2:3:end); % 2314 long
speeds = k(3:3:end); % 2314 long
whos k
whos lats
whos lons
whos speeds

Name Size Bytes Class Attributes

k 6943x1 55544 double

Name Size Bytes Class Attributes

lats 2315x1 18520 double

Name Size Bytes Class Attributes

lons 2314x1 18512 double

Name Size Bytes Class Attributes

speeds 2314x1 18512 double

As you can see, k is not a multiple of 3 so lats is one element longer than the other two. Why is that?

Moreover, the wind speeds are practically the same value as lats and lons (they are all around values 0-8), which is suspicious unless you measured the wind near the north pole. Please attach the actual data.

[MAT NIZAM UTI]

https://drive.google.com/drive/folders/1tFOl0ZHQo4XzB-VGvi-LBLPg_lERG98u?usp=sharing The reason why the wind speeds value are approximately arround 0-8 because the location is at the equator region and generally the equator region recieved less wind compare with northern and southern hemisphere. Here i attach the actual data.

[Image Analyst]

But what about this question that you didn't answer:

As you can see, k is not a multiple of 3 so lats is one element longer than the other two. Why is that?

[MAT NIZAM UTI]

I have tried run this coding

% Read in data

%k = xlsread('ACTUAL DATA_WIND SPEED.csv');

k = xlsread('AVERAGE_WINDSPEEDS_1.csv');

lats = k(1:end,1); % 2315 long

lons = k(2:end,2); % 2314 long

speeds = k(3:end,3); % 2314 long

whos k

whos lats

whos lons

whos speeds

But, when I compare the values of lats, lons and speeds with the actual data. I get this

1) The number of elements of lats (or LATITUDE_AFTER READ) is the same with the actual latitude.

2) But for lons (or LONGITUDE_AFTER READ) and speeds (or WIND SPEEDS_AFTER READ), there were differences in terms values and number of elements compared to the actual longitude and wind speeds.

As you can see too, at the end of the data for each column, the number of elements for lons and speeds is not same with the actual, thus may be this is the reason why lats is longer than lons and speeds.

[Image Analyst]

So can we just take the first 2314 values and ignore the extra lat?

[MAT NIZAM UTI]

Sure..well I dont really know how the matlab works, because after comparing the actual values and the after read values, both lons and speeds were different with the actual data.

https://drive.google.com/drive/folders/1tFOl0ZHQo4XzB-VGvi-LBLPg_lERG98u (this is my very actual data) Column C until LB is the wind speeds values.

Answer 2

[H R]

Plese see: https://www.mathworks.com/help/stats/kmeans.html.

If your data is in a matrix format X, then you can use the following:

[idx,C] = kmeans(X,3,'Distance','cityblock','Replicates',5);

[H R]

Make sure to use a correct 'Distance' that makes sene to you. Try different distances and observe the results.

[MAT NIZAM UTI]

My wind speed data is in grid format which is the data were arranged based on 0.25 degree for each latitude and logitude and each point is provided with one value of wind speed, so in your opinion what is the suitable "Distance" that I should use to cluster the wind speed data.

[H R]

It seems you basically have v=f(x,y). So, I don't think you can mix dependent and independent variables in clustering i.e. (x,y,v) to gain useful information (if this is the case). I think it's better to use a supervised method instead. Alternatively, you may try to perform clustering using (x,y) and then see if the outcome of the clustering can be useful to give information about v. In doing so, I guess Euclidean distance would be enough. It seems there is a relevant paper on your subject as well: https://ieeexplore.ieee.org/document/7884477.

[MAT NIZAM UTI]

Let say, I remove the Lat Long data and just used the wind speed data and arrange the wind speed data into one matrix (100:1), is it possible ?

[H R]

Yes, every thing is possible (even using 1D data) , but you have to finally check what you are looking for from the clustering task and check if the outcome makes sense to you.

[MAT NIZAM UTI]

Here is my coding, and I have an error on it

Error using horzcat

Dimensions of matrices being concatenated are not consistent.

Error in k_mean (line 7)

idx1=kmeans([X1 Y1],numClusters,'Replicates',5);

This is the code:

k = xlsread('WIND_26YEARS.csv');
X1=(1:6943);
Y1=randn(6943,1); 
numClusters=3;
idx1=kmeans([X1 Y1],numClusters,'Replicates',5);
pointclust=repmat(idx1,1,numClusters)==repmat(1:numClusters,numel(idx1),1);
colors=hsv(numClusters);  
for j=1:numClusters,
    plot(X1(pointclust(:,j)),Y1(pointclust(:,j)),'Color',colors(j,:));
    if j==1,
        hold on;
    end;
end,
hold off;