Neuromongers, don't pass by :) need advice - page 14
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
No, now there is an urgent need to do another project, neuronics has been put on hold for a while.
The goal is to normalise quotes relative to volatility.
Hello Andrey. Sorry for the long absence. I see that you have obtained quite good results with the echo network. I tried to join the discussion in your topic, but I've found nothing smart to say. I don't know if devolatilization will help you or not (most likely not), but you definitely need normalization of inputs if your network neurons are non-linear. By the way, about non-linearity. It can be set in two ways. (1) The output of a neuron saturates at large inputs tending to -1 or 1 (hyperbolically tangent). (2) The output of a neuron is described by an exponential function with a certain threshold. Most network designers choose the first function. But brain neurons use the second one. I do not know whether it helps you or not.
I myself have stopped believing in price predictors. I think it's a dead end. I'm more interested in buy/sell type classifiers now. Some will argue that it is the same predictor - when it gives a buy signal it predicts that price will go up. It does not matter. The human brain is a classifier, not a predictor. And it uses several neural layers for non-linear transformation of input information. It is this non-linear transformation that interests me more than classification. Classification can be done by perceptron, SVM, kNN, or any other known method.
I myself have stopped believing in price predictors. I think it's a dead end. I'm more interested in buy/sell type classifiers now. Some will argue that this is the same predictor - when it gives a buy signal it predicts that price will go up. It does not matter. The human brain is a classifier, not a predictor. And it uses several neural layers for non-linear transformation of input information. It is this non-linear transformation that interests me more than classification. Classification can be done by perceptron, SVM, kNN, or any other known method.
As I understand it, this conclusion is indeed drawn by many on the application of NS.
Here, if someone doesn't have, I'm putting up a very interesting dissertation on this topic, there is a lot of digging over the material.
if you look at the conclusion of the thesis straight away - it was the approach to the network as a classifier of market situations that gave the best approach, and all the others stalled.
People manually classified situations on a training sample and then the NS was trained to recognise these situations, similar to how the NS is trained to recognise images - this gave the best results.
I haven't used neural networks yet, but I don't have much faith in predictions.
But if I could estimate the volume of all trades separately for the broker - buy and sell, I think it would be useful ))))