Re: That's It


"Mike" <Mike@xxxxxxxxxxxxxxxxxxxx> wrote in message news:WkmVl.27991$c45.13769@xxxxxxxxxxxxxxxxxxxxxxx
On Wed, 03 Jun 2009 00:19:02 +0000, Bill Reid wrote:
"Mike" <Mike@xxxxxxxxxxxxxxxxxxxx> wrote in message
news:DlcVl.27954$c45.5665@xxxxxxxxxxxxxxxxxxxxxxx
On Mon, 01 Jun 2009 22:04:45 +0000, Chippy wrote:

It's software that teaches itself and figures out what is "real" and
what isn't, and what's most probably going to happen, before it
happens.

"teaches itself" ?

i've always been intrigued by the notion of a computer being able to
"learn" but have yet to see one do anything other than exactly what it
was programmed to do.

"Neural networks" are computer programs that do indeed "train"
themselves to discover repeating patterns in data sets...you just feed
in the data, typically several "time series" of some sort, and they
automagically find the repeating correlations in the data and then can
make a "prediction" about the future results of a target data set...

To make market predictions, you get a bunch of time series of, say, P/E
ratios, dividend yields, inflation rates, money supply growth, etc.,
whatever you think might influence the market, and a time series of
market prices, feed it all into the neural network software, and a
little while later it will give you "predicted" future prices for the
stock market.

so if i extend the trendline on my sp500 log chart (which has data from
1926 to 2008) to 2010 & beyond can i say i've "trained" my computer to
"discover repeating patterns in data sets" and "predicted future prices
for the stock market" ?

If YOU do it, it's "natural intelligence" (or stupidity), if a COMPUTER
does it, it's "artificial intelligence" (or stupi...no wait, it's not fair to call
a retarded machine "stupid").

Or at least, that's the dream, a dream that has been around for at least
30 years now...there are a few advisory services that claim they base
their advice on the output of neural networks...

i would've thought if computers could actually learn they would have
long since figured out how to do basic stuff like, oh i dunno know,
maybe find optimal solutions to random configurations of rubiks cube
with sub-second response time (ie. not needing massive amounts of
processing & data to do it).

They can't? I dunno, don't really want to think about it, might make my
head hurt...

i'm thinking that "solving" the stock market is way more complicated
than a simple child's toy...

Uh, yeah, because the toy is dealing with a fixed finite CLOSED set of
probabilities...the market is prone to what are called "exogenous
shocks" (but this may actually be a misnomer, it's just that it SEEMS
that things seem to "come from nowhere" and dramatically change your
tidy theory of market influences just when you think you've got it all
"figured out").

And oddly enough, that's the problem with neural networks as
well..."neural networks" are part of a branch of computer science known
as "artificial intelligence", and the dirty secret of this technology is
that there IS such a thing as "artificial stupidity"...

well then i won't hold my breath on computers being able to predict the
stock market until they've at least mastered a simple child's toy (as in
actually learning how to do it as opposed to doing little more than using
heuristics to mildly improve brute-force searching and calling it
"artificial intelligence")

Hmmm, gotta give you credit for using terms like "heuristics" and "brute
force", too bad the OP is unable to converse using those terms. I tend
to think of neural networks as being more "brute force" rather than
"heuristics" because they tend to use digital "scoring" of what are
essentially automated combinatorial Bayesian chains, but whatever...

In any event, I'm not as skeptical of computer predictions of the
stock market as you apparently are, and particularly not very skeptical
of "neural networks". I DID configure a neural network to make
stock market predictions over 20 years ago, and have always
described the results as "interesting". Every time the topic comes
up again, I have an urge to blow the dust off my neural network
software and try it with the data sets I am currently using for the
software I wrote, just to see what would happen.

And you really shouldn't be so disparaging of "brute force". As
computers become exponentially more and more powerful and
faster, "brute force" appears more and more like instantaneous
magic. I strongly believe that it doesn't matter how you get to
the answer, as long as it is the CORRECT answer and you can
get to it in time to make a practical "real-life" decision...

For a perfect real-life example of this, note carefully the
design and progression of computer chess programs...and for
a REAL philosophical "thought experiment" about "artificial
intelligence", ponder the concept of "wisdom", why is it that
people tend to become more "wise" as they age even though
you can be sure that their brains are detiorating along with
their bodies...

---
William Ernest Reid
Post count: more than my chronological age

.

Relevant Pages

  • Re: Thats It
    ... To make market predictions, you get a bunch of time series of, say, ... for the stock market"? ... I'm not as skeptical of computer predictions of the stock ... I DID configure a neural network to make stock ...
    (misc.invest.stocks)
  • Re: How low will the stock market go ?
    ... I avoid stock market predictions for the same reason most people do. ... make a profit if it were not for the American taxpayer and consumer? ... We can't hang the corporations, ...
    (soc.retirement)
  • Re: How low will the stock market go ?
    ... I avoid stock market predictions for the same reason most people do. ... Not according to European observers. ... Europe for cheap vacations. ...
    (soc.retirement)
  • Re: Thats It
    ... Mr Reid is kinda cranky... ... I'm sorry you were a failure at this, but my design is my own and unrelated to any neural network out there, as I have solved some of the problems others have not. ... Yeah, this nation has never seen a stock market crash, bank failures, ...
    (misc.invest.stocks)
  • Re: Thats It
    ... I know NOTHING about such programing or what it takes to make it work. ... I'm sorry you were a failure at this, but my design is my own and unrelated to any neural network out there, as I have solved some of the problems others have not. ... Yeah, this nation has never seen a stock market crash, bank failures, ...
    (misc.invest.stocks)