Thread Subject: Seeking Simple HMM Example(s)

I am trying to teach myself Hidden Markov Models [HMMs]. I
am seeking some simple examples to understand how and why
they are used.

I realize that this is a strange request, instead of begging
for code, I'm effectively asking for data. (A good
discussion would do to.)

In a Hidden Markov Model, a measurement is OBSERVED, and
this hints at the true STATE of an underlying variable.
Supposedly, the STATE is hidden from you.

The problem is that HMMs are usually explained in terms of
flipping coins. The coins are fair, and there is nothing
"hidden" about them, so the examples are wasted on me. If I
OBSERVE that the coin is a head-up, then the STATE is head-up.

There is another common example, colored balls in urns.
Again, there seems no possibility for error in the
measurement. You look at the ball, you know the color. Why
would you need a model?

Hidden Markov Models are important because they form the
basis for many other graphical techniques for Pattern
Classification -- Bayes Nets, Factor Graphs, Markov Fields,
etc...

Here is one slightly better example. Suppose you were
trying to figure out information about a road. You want to
know if it is a small street or a large highway. So you
observe the vehicles that go by -- some are cars, some are
trucks. In this way you are observing something which is
not the actual state of the road. However, I would never
use an HMM for this.

Another example might be if you were looking through foggy
glass, or your camera was out of focus. The examples I seek
must have the possibility that what you observed might be
the wrong state of affairs, and it must be applicable for
HMMs, so that I can program it up and understand it.

Thanks for any help. Best Regards...

"Pat Finder" <pfinder.nospam@netacc.net> wrote in message
<fsb812$aep$1@fred.mathworks.com>...
> I am trying to teach myself Hidden Markov Models [HMMs].
I
> am seeking some simple examples to understand how and why
> they are used.
>
> I realize that this is a strange request, instead of
begging
> for code, I'm effectively asking for data. (A good
> discussion would do to.)
>
> In a Hidden Markov Model, a measurement is OBSERVED, and
> this hints at the true STATE of an underlying variable.
> Supposedly, the STATE is hidden from you.
>
> The problem is that HMMs are usually explained in terms of
> flipping coins. The coins are fair, and there is nothing
> "hidden" about them, so the examples are wasted on me.
If I
> OBSERVE that the coin is a head-up, then the STATE is
head-up.
>
> There is another common example, colored balls in urns.
> Again, there seems no possibility for error in the
> measurement. You look at the ball, you know the color.
Why
> would you need a model?
>
> Hidden Markov Models are important because they form the
> basis for many other graphical techniques for Pattern
> Classification -- Bayes Nets, Factor Graphs, Markov
Fields,
> etc...
>
> Here is one slightly better example. Suppose you were
> trying to figure out information about a road. You want
to
> know if it is a small street or a large highway. So you
> observe the vehicles that go by -- some are cars, some are
> trucks. In this way you are observing something which is
> not the actual state of the road. However, I would never
> use an HMM for this.
>
> Another example might be if you were looking through foggy
> glass, or your camera was out of focus. The examples I
seek
> must have the possibility that what you observed might be
> the wrong state of affairs, and it must be applicable for
> HMMs, so that I can program it up and understand it.
>
> Thanks for any help. Best Regards...

The analogy that I liked best was to assume you were living
in a cave, and had with you a hydrometer. You couldn't see
the outside world, but could see the readings of the
hydrometer. The underlying variable was whether or not it
was raining on the outside of the cave, what you could
observe was the moisture content of the air in the cave.

Not quite what you were after, but I hope it helps

Regards

Dave Robinson