Okay, assume we have a very fast neural network that can analyze x (where x can be large) amounts of data very fast, ready to trade the coming set of bars (only thinking 1-3 bars, not more).
What would be the best predictive algo to feed such neural network with?
"Perceptron" (various implementations online) is a reasonably good one, but there is likely better. Which one do you think is best? Why?
Not financial advice. For testing purposes only. Do not use on live account
If the best prdictive algo could exist - forex traders could be the richest people in the world)) Algo's are good, but u need to use it in exact moment!
Okay, assume we have a very fast neural network that can analyze x (where x can be large) amounts of data very fast, ready to trade the coming set of bars (only thinking 1-3 bars, not more). What would be the best predictive algo to feed such neural network with? "Perceptron" (various implementations online) is a reasonably good one, but there is likely better. Which one do you think is best? Why?
i played around with those 3 or 4 years ago, as far as i remember it didn't lead anywhere.
unless you have a super computer to work with a bigger data sample and higher number of Neurones i doubt you can get any viable result.
{quote} Very nice! Reading up on Wavelets. If you have any work done on this already, shoot me a PM and we can join forces.
Ignored
This is on my to do list.Wavelets, kernels few more things.I find this article interesting too https://www.mql5.com/en/articles/412 it looks interesting, what do you think?
Joined Jan 2007
|
Status: developing...
|974 Posts
@roel13
The premise of the thread is that there is a single best prediction method. This contradicts a statistical learning theory known as the "no free lunch theorem." This theorem essentially states that there is no universal best method for all applications.
It turns out that there are many ways to do so, but no best one.
Quote
Disliked
This lack of universally best choice can actually be formalized in what is
called the No Free Lunch theorem, which in essence says that, if there is no
assumption on how the past (i.e. training data) is related to the future (i.e. test
data), prediction is impossible. Even more, if there is no a priori restriction on
the possible phenomena that are expected, it is impossible to generalize and
there is thus no better algorithm (any algorithm would be beaten by another
one on some phenomenon).
Quote
Disliked
B No Free Lunch
We can now give a formal definition of consistency and state the core results
about the impossibility of universally good algorithms.
The answer as to the "best" depends wholly on how you define the specific problem. Sorry, there is no free lunch here. Theory doesn't point to an answer of "best", so this will need to be by the old fashioned method: trial and error.
