What to feed to the input of the neural network? Your ideas... - page 28
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
train two grids - one in buy only, one in sell.
switch on both :-)
Original!
It's original!
there is also a correction (later, you had time to reply before I could add it)
"then add a network (or just alg.) collision resolution - so that at the same time in different directions do not trade".
the third network is more fun, more practical, more graphical :-))
there's a correction (later, you had time to reply before I could add it)
"then add a network (or just alg.) collision resolution - so that at the same time in different directions do not trade"
the third network is more fun, more practical, more graphical :-))
Thanks for the idea. It looks a bit complicated, as I still need to understand what a collision resolution network is)))
If you also have an idea how to solve the problem locally and concisely, within one simple network - please, write too
It's original!
They used to put up grails like this:
i.e. the future quote was peeked at.
and based on the known outcome of the trade.
Maybe the same method could be used to train a neuron?
Maybe it will wise up at last? ;)
Thanks for the idea. It looks a bit complicated, as I still need to understand what a collision resolution network is )))
If there is also an idea how to solve the problem locally and concisely, within one simple network - please post too
e.g. start with a simple - stupid algorithm "if both networks are signalling in different directions at the same time, nobody gets anything".
then expand the notion of "simultaneously" and start regulating "who-what", i.e. volumes/lots.
something like that.
they used to put out grails like this:
i.e., future quotes were peeked at.
and by the known outcome of the trade.
could the same method be used to train neuronics?
Maybe it will finally wise up? ;)
on the contrary :-)
constant peeking into the future ruins the whole Neuro-Deep-Machine Learning application.
As long as no one can calculate the future result with more than "finger-in-the-sky" accuracy from a numerical series, all NN's endeavours are meaningless.
without a peek into the future, nobody knows what to find.
I have not quite correctly highlighted the problem:
Neuron can trade back and forth, even the most primitive one (signal/no signal, up/down).
And the optimiser just gives out sets in the tops, among which there are non-stop BUY-SELL pounding. They are uninteresting, clumsy and crude.
The problem is that there are no BUY-SELL sets in the tops that have breaks between these trades on the chart. But at the same time, the very tops have sets that have breaks between SELL trades. And if the chart was bullish, there are breaks between BUYs. But at the same time, not a single SELL trade to remove corrections in +. And the optimisation is set to "Maximum Balance". Here it is, the correction - take the correction in the plus! But no. Either it will wait it out, or SELL will open on it, or BUY-SELL will open, without breaks.
Here is the top set from the "Maximum Balance" mode.
Although in idea the optimiser should set the weights so that either to pound on extrema (near them), or to make breaks up to extrema and open an opposite deal on them or a little later.
But no.
Maybe purely mathematically it is impossible (to grail both ways), because the weights are sharpened for a "yes/no" signal, i.e. "trade down/not trade down", "trade up/not trade up", and not "trade there/now here"
The funny thing is that the picture shows a neuron from the article about neural network. It has 3 layers of 5 neurons and another 3 are outputs. Even it can't make a set that would have breaks between multidirectional trades, but this set would be in the top (which is logical, because that's when the maximum profit is achieved).
UPD
I'm not talking about super-duper models, onyx, python, etc. I'm talking about the simplest MLP mechanisms.
I didn't quite cover the problem correctly:
A neuron can trade back and forth, even the most primitive one (there is a signal/no signal, up/down).
And the optimiser just gives out sets in the tops, among which there are non-stop BUY-SELL pounding. They are uninteresting, clumsy and rough.
The problem is that there are no BUY-SELL sets in the tops that have breaks between these trades on the chart. But at the same time, the very tops have sets that have breaks between SELL trades. And if the chart was bullish, there are breaks between BUYs. But at the same time, not a single SELL trade to remove corrections in +. And the optimisation is set to "Maximum Balance". Here it is, the correction - take the correction in the plus! But no. Either it will wait it out, or SELL will be opened on it, or BUY-SELL will be opened without interruptions.
Here is the top set from the "Maximum Balance" mode.
Although in idea the optimiser should set the weights so that either to pound on extrema (near them), or to make breaks to extrema and on them or a little later to open with the opposite deal.
But no.
Maybe purely mathematically it is impossible (to grail both ways), because the weights are sharpened for a "yes/no" signal, i.e. "trade down/not trade down", "trade up/not trade up", and not "trade there/now here"
The funny thing is that the picture shows a neuron from the article about neural network. It has 3 layers of 5 neurons and another 3 are outputs. Even it can't make a set with breaks between multidirectional trades, but it can't make a set to be in the top (which is logical).
UPD
I'm not talking now about your super-duper models, onyx, python, etc. I'm talking about the simplest MLP mechanisms.
So you've been taught the wrong thing by NN.
it is not the fact of profit that matters, but the value of %profit/time_in_market.
Ivan Butko #:
And if the chart was bullish, there were breaks between BUYs. But at the same time, not a single SELL to remove corrections in +. And the optimisation is "Maximum balance". Here it is, the correction - take the correction in the plus! But no. Either it will wait it out, or SELL will be opened on it, or BUY-SELL will be opened without interruptions.
Ivan, hello. It seems to me that the network itself will not split the chart into a trend and a pullback. Do it yourself, and then one network works in a trend and the other in a pullback.
Ivan, hello. It seems to me that the network itself will not split the chart into a trend and a pullback. Do it yourself, and then one network works in trend, the other in pullback.