KMEANS help!!!!!!!!!!!
Shweta
I am using Matlab's KMEANS to cluster data. I find that when I call
the function each time I run the code it clusters differnently, as in
the same points are put in different clusters every time the function
is called. Do any of you happen to know why ?
Any inputs on this are highly appreciated...
Thanks
Shweta
Dan Pearce
Tom Lane
> the function each time I run the code it clusters differnently, as in
> the same points are put in different clusters every time the function
> is called. Do any of you happen to know why ?
Shweta, the answer is here:
KMEANS K-means clustering.
...
[ ... ] = KMEANS(..., 'PARAM1',val1, 'PARAM2',val2, ...) allows you to
...
'Start' - Method used to choose initial cluster centroid positions,
sometimes known as "seeds". Choices are:
{'sample'} - Select K observations from X at random
The algorithm needs a starting value, which in this case is a set of K
univariate cluster means. I suspect if you fix the starting values, your
results won't vary.
There's another option:
'Replicates' - Number of times to repeat the clustering, each with a
new set of initial centroids [ positive integer | {1}]
This takes advantage of the random starting points by running the procedure
multiple times, and taking as a final answer the results that minimize the
within-cluster distances.
-- Tom
Peter Perkins
> The algorithm needs a starting value, which in this case is a set of K
> univariate cluster means. I suspect if you fix the starting values, your
> results won't vary.
It may also be that the different results on different runs are simply
permutations of the cluster numbers.
salih tuna
i am also using kmeans clustering in matlab. i was
wondering if it is possible to manually choose the initial
points of the clusters? if yes how do i need to enter it?
thanks
salih
Shweta <shwetaan...@gmail.com> wrote in message
<ef511...@webcrossing.raydaftYaTP>...
Tom Lane
> wondering if it is possible to manually choose the initial
> points of the clusters? if yes how do i need to enter it?
The answer is here:
>> help kmeans
KMEANS K-means clustering.
IDX = KMEANS(X, K) partitions the points in the N-by-P data matrix
...
'Start' - Method used to choose initial cluster centroid positions,
sometimes known as "seeds". Choices are:
...
matrix - A K-by-P matrix of starting locations. In this case,
you can pass in [] for K, and KMEANS infers K from
the first dimension of the matrix. You can also
supply a 3D array, implying a value for 'Replicates'
from the array's third dimension.
-- Tom
pires
<fdej36$1hc$1...@fred.mathworks.com>...
How to set???
No matter how i configure...Also failed. can u provide me
example...
Tom Lane
news:fts83v$b63$1...@fred.mathworks.com...
> No matter how i configure...Also failed. can u provide me
> example...
Here I use kmeans on the Fisher iris data, starting with the three species
means. I plot the results and compare the clusters to the known species.
-- Tom
>> load fisheriris
>> m = grpstats(meas,species)
m =
5.0060 3.4280 1.4620 0.2460
5.9360 2.7700 4.2600 1.3260
6.5880 2.9740 5.5520 2.0260
>> g = kmeans(meas,3,'start',m);
>> gplotmatrix(meas,[],g)
>> crosstab(g,species)
ans =
50 0 0
0 48 14
0 2 36
Kesede
I executed his example and I obtained the same answer that I have been obtaining in my application, or be:
??? In an assignment A(I) = B, the number of elements in B and
I must be the same.
Error in ==> prwaitbar at 159
N(DEPTH) = n; % complete the admin
Error in ==> kmeans at 57
prwaitbar(maxit,'k-means clustering');
I already verified the code of the prwaitbar and I didn't get to solve the problem. Would know how to tell me how it could avoid this mistake?
Thank you very much!!
Pekka Kumpulainen
Are you sure you are using the kmeans provided in the stats toolbox? Type
which kmeans
In my installations 2009a or 2010a the stats toolbox versions of kmeans don't have use anything called prwaitbar.
Kesede
I typed,
>> which kmeans
C:\Matlab\toolbox\stats\kmeans.m
my version is 7.0.0.19920 (R14), what could be?
Tom Lane
>> > obtaining in my application, or be:
>> >
>> > ??? In an assignment A(I) = B, the number of elements in B and
>> > I must be the same.
>> >
>> > Error in ==> prwaitbar at 159
>> > N(DEPTH) = n; % complete the admin
>> >
>> > Error in ==> kmeans at 57
>> > prwaitbar(maxit,'k-means clustering');
> I typed,
>>> which kmeans
> C:\Matlab\toolbox\stats\kmeans.m
There is a non-MathWorks thing called prtools. I don't know what it is, but
it has a kmeans function, a crossval function, and a prwaitbar function. It
appears you were calling the prtools function originally. But then later,
when you issued the "which" command, you didn't see that. I can't explain
the difference.
If you give the command
which kmeans -all
you may see both the Statistics Toolbox version and the prtools version. If
so, try temporarily removing prtools from your path. (Or just cd out of that
directory if that's where you are.) If you just see the statistics one
again, try repeating the example to see if it works this time.
-- Tom
Kesede
yafit maymon
[g ,sumd]= kmeans(meas,3,'start',m); - it aill return the centers in sumd but i would like it to return the sum of distance for each cluster ...
can you help me with that ??
thanks
yafit
Tom Lane
> hi u have also used the kmeans function with starting points matrix , as
> your exemple shows , but i have noticed that it wont allowe me to add more
> values for the function , for example :
> [g ,sumd]= kmeans(meas,3,'start',m); - it aill return the centers in sumd
> but i would like it to return the sum of distance for each cluster ...
> can you help me with that ??
If you type "help kmeans" you will see:
[IDX, C, SUMD] = KMEANS(X, K)
You have named your second output variable sumd, so I suspect you really
want the third output. Here's one way to do that:
[g, ~, sumd] = kmeans(...)
-- Tom
yafit maymon
i though that was the problem , only didnt know how to solve it, your soulution seems simple i wish it worked, for some reason it doest , i keep getting error:
(x has data, m has 3 centers if i run it without the sign it work perfectly-can you see anything wrong???
>> [idx,~,sumd] = kmeans(x,3,'start',m);
??? [idx,~,sumd] = kmeans(x,3,'start',m);
|
Error: Expression or statement is incorrect--possibly unbalanced (, {, or [.
thank you!
Steven_Lord
"yafit maymon" <yafi...@gmail.com> wrote in message
news:ia4fn2$30t$1...@fred.mathworks.com...
If you're using a version of MATLAB prior to release R2009b, this syntax
will not work for you.
http://www.mathworks.com/help/techdoc/rn/br5fo8o-1.html#br65zmd-1
Replace the tilde in the output argument with a dummy variable name.
--
Steve Lord
sl...@mathworks.com
comp.soft-sys.matlab (CSSM) FAQ: http://matlabwiki.mathworks.com/MATLAB_FAQ
To contact Technical Support use the Contact Us link on
http://www.mathworks.com
yafit maymon
hi steve,
so my 2008 version is the problem ..
i tried your solution and it works perfectly
thak you !
yafit
Raj Kodirekka
i think you are very familiar with MATLAB kmeans, i am trying to rewrite this MATLAB kmeans for different distance measure.
as you know it has got 2 phases, Batch phase and online phase.
i know what is happening in Batch phase but i am not clear about online phase so can you please explain it in detail? it will be a great help to me.
thanks in advance.