[3dem] clustering algorithms

[Dan Asarnow]

Hi,
Just FYI, the classifications in cryoSPARC *do* report membership
probabilities for every class. They take the maximum probability class
(apply a threshold, default 0.9) during particle selection, but all the
values are present in the metadata file.

Best,
-da

On Thu, Aug 31, 2017 at 7:59 AM, Penczek, Pawel A <
Pawel.A.Penczek at uth.tmc.edu> wrote:

> Hi,
>
> I may say I am somewhat familiar with the technique and I even published
> on the subject in a distant past:
> Leszczynski, K., Penczek, P. and Grochulski, W.: Sugeno's fuzzy measure
> and fuzzy clustering. Fuzzy Sets and Systems 15:147-158, 1985.
> There are also some comments in the supplement of:
> Cheng, Y., Grigorieff, N., Penczek, P.A., Walz, T.: A Primer to
> Single-Particle Cryo-Electron Microscopy. Cell, 161:438-449, 2015.
>
> Regrettably, it is not a subject that easily yields itself to a discussion
> on a mailing list.  More like 30 minutes lecture.
>
> Very briefly, fuzzy clustering does not merge anything with maximum
> likelihood (ML) approach.  To the contrary, fuzzy sets theory was
> developed in 1962 by Zadeh as a substitute for probability-based
> descriptions.  The idea was that having
> arbitrary “membership functions” designed by the researcher instead of
> probability distributions
> derived from the data would give more “freedom" and improve “accuracy”,
> particularly for small samples.
>
> There was a surge of fuzzy sets-based clustering algorithm developed in
> the 80s (see above).
>
> I will not get into detailed comparative analysis of ML-based versus fuzzy
> clustering methods here.
> Very briefly, ML is not really applicable to clustering unless some
> heuristics are added.  Fuzzy methods are all heuristics.
>
> Greetings from Houston.
> We are slowly getting back to normal and the fact I could write the above
> note means my senses are coming back,
> but it is an arduous process.
> Pawel.
>
> > On Aug 31, 2017, at 9:24 AM, Morgan, David Gene <dagmorga at indiana.edu>
> wrote:
> >
> > Hi,
> >
> >    The recent flurry of e-mail about k-means clustering has made me
> wonder whether anyone in our field has tried to use c-means clustering
> instead.  As I understand it, c-means clustering is an application of
> "fuzzy logic" to the clustering problem, and another way of describing it
> (one that might spark a bit more interest) would be to say it merges a
> clustering algorithm with maximum likelihood:  at the end of the process,
> every particle has a weighted membership in every class.  I have no idea
> whether this would actually be useful for our problems, but I can see some
> ways that it might be.
> >
> >    So, has anyone tried it, and if so, what are the conclusions?  If
> no-one has tried it, maybe someone will!
> >
> >    Finally, best wishes to our friends in the Houston area.
> >
> > --
> >            David Gene Morgan
> >        Electron Microscopy Center
> >             047D Simon Hall
> >             IU Bloomington
> >          812 856 1457 (office)
> >          812 856 3221 (3200)
> >      https://urldefense.proofpoint.com/v2/url?u=http-3A__
> iubemcenter.indiana.edu&d=DwICAg&c=6vgNTiRn9_
> pqCD9hKx9JgXN1VapJQ8JVoF8oWH1AgfQ&r=vDDf9rsFxPMXm8JgJa6hc4B9V4qKr7
> wftnDkLIRdshI&m=fEKKv4cgzqj-A0nGXzOoPRMCjRmNIJYv0_YA97v_cBc&s=
> QNvEESgEl2FMMtbbpVdweUr7o-kkUga-fydybPMps4o&e=
> > _______________________________________________
> > 3dem mailing list
> > 3dem at ncmir.ucsd.edu
> > https://urldefense.proofpoint.com/v2/url?u=https-3A__mail.
> ncmir.ucsd.edu_mailman_listinfo_3dem&d=DwICAg&c=6vgNTiRn9_
> pqCD9hKx9JgXN1VapJQ8JVoF8oWH1AgfQ&r=vDDf9rsFxPMXm8JgJa6hc4B9V4qKr7
> wftnDkLIRdshI&m=fEKKv4cgzqj-A0nGXzOoPRMCjRmNIJYv0_YA97v_cBc&s=
> TCybivSDZyApAlOEf5D4pouWuFZNhTjkBylAeSdnS4Y&e=
>
> _______________________________________________
> 3dem mailing list
> 3dem at ncmir.ucsd.edu
> https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20170831/ab8f4ce6/attachment-0001.html>