Local min, max, nearest neighbour

Find local minima, maxima, match value in vector.

Description

Set of functions to look for local minima or maxima in a vector, find a point inside a vector closest to a certain value and small function to clear everything.

• Acknowledgements: This submission has inspired the following:; Plot cylinder
• MATLAB release: MATLAB 6.5 (R13)

[Sergei Koptenko]

Appology to all who loaded this file!
1) one bug was found and fixed in lmax_pw.m and lmin_pw.m
2) match.m wasn't stable dealing when first and last points were in the range- Fixed
Please reload zipfile in two-three days (new name should be "local_ext_mod.zip")

[Nick Disabato]

I'm using this to find minima in a plot which involves a lot of sharp downward spikes; the goal is to find at what points these spikes land. It's a great program, but a future improvement may be to add in code that will register a minima if it attains a certain percent difference from a group of points previous to it. This seems sort of eccentric, but it may help in applied mathematical applications

[David Liu]

I guess there is a bug when the "peak" is flat, such as a=[3 4 5 5 4 3]. lmax returns empty.

[Sergei Koptenko]

Not quite a bug but a feature :-)
If you go into the function body there are four lines commented out NN 41, 42, 45 and 46 that define if flat spots should be included in results. If you un-comment them, function will give a correct answer.

[Bing Ouyang]

Here is another example that it did not report correct answer.

If the input has two flat steps:
       a=[0 1 1 2 2 0]
The middle step will be considered a local maxima...
indd =[2 4]

[Sergei Koptenko]

1) another example? I?m sorry Bing, but if you just try to implement measures I mentioned in the first message you will see that function indeed gives the correct answer to the first example.
2) Concerning your example, the correctness of the answer lay not in the function?s output but in your original assumptions. The problem here is how to define the local maxima for a long flat spot. For instance, in [0 0 1 1 0 0] position of maximum can be defined at 3rd OR at the 4th position, at your liberty. In vector [0 0 1 1 2 2 1], 0011 can be defined as one-sided local maximum (a ledge) or as an absence of the maximum, depending on your views and needs. This is why I chose to adhere to the strict maxima criteria excluding flat spots altogether, but giving the possibility of including them if such a need arises through commented lines in the code.

[Dragos Maciuca]

 

[Simon de Leon]

minor bug:
returns empty matrices for
[a,b]=localmax([0 0 0 0 2 0])

fix:
in max/min search loop, change
--> while i < len_x-1
to
-->while i < len_x

Other than that, does exactly as described.

[Daniel C]

it's very good

[Leon King]

great. works like a charm

[Felix TTTTTTTTT]

[a,b]=localmax([0 2 0 0 0 0]) also has to tuned by hand. Nie code, but some rough spots. Could be improved.

[Rajesh Krishnan]

Incredibly useful!!! lmax identifies some local minima points if the number of passes are low.

[Withawat W.]

Thanks a lot! You save my time :)

[Kianoush Nazarpour]

Thank you! You make the life easier!

[vamsee somavarapu]

nice work and very helpful

[Sergei Koptenko]

Steven Kay ([e-mail]) is a fictional character:
....If you do so, please include this problem report. You can delete your own text from the message returned below.

The Postfix program

<[e-mail]>: host cast.navnet.net[64.5.195.2] said: 550
    <[e-mail]>: User unknown in local recipient table (in reply to
    RCPT TO command)....

[Niranjan Prasad]

The life became easier now.

[jon erickson]

Works great...a real time saver.
Another user commented a while back that he is using it for "spike" detection. I am doing the same. If you want to find the spikes only, and not all local minima, first use "find" command to locate all points that have amplitudes above a certain criteria--all points with amplitude 5 sigma above the rms value of the signal, for instance

[Cynthia G.]

i find extreme.m more simple to use

[leila ayoubian]

 

[Hyeongyong Jeon]

Very good! thanks for sharing us~!

[L Lu]

 

[J D]

I love you! This has made my life so much easier

[fugazi p]

Interesting. How can this help me tackle my problem, here: http://www.mathworks.fr/matlabcentral/newsreader/view_thread/165101

Comments and help appreciated.

[B B]

This is very helpfull! Thx! A LOT!

[o l]

 

[Ron Sostaric]

I used the "match" function. Exactly what I was looking for, and works very well.

[Claudio P.]

A couple of comments:
1) help for lmin is wrong, should be [b,a]=lmin... rather than [a,b]=lmin...
2) lmax is not finding the exact max with high 'filt' values, I had to increase the rng value at lmax.m line 55 in oder to have it wokring fine (maybe rng should not have a fixed value but should be proportional to filt?)

apart from that, everything ok ;)

[R Jermia]

Excellent!! It rocks.

[Sean Vitousek]

Nice! Works quite well.

[Robert Hovden]

Great, simple program. Works well, good examples. This program will save you time since there is no need for you to write what Koptenko has already done well.

[levent yüksek]

Usefull program saved my time. Thanx

[Peter Lo]

 

[Peter Lo]

Nice work. It save a lot of my time.
Two comments:
1) Would like to see you add a threshold value so that it only find min and max that are below or above the threshold. I made a simple modification to the lmax and lmin to include that
2) Would also like to see adding the filtered curve to plot with the graph. I see the same problem with Claudio P. that in a high noise situation, the maxima are different for low and values of filt. Adding the smoothed curve will help me to determine which filt value to use

[Idillus]

Nice ans usefull

[Q L]

Thank you so much.