Machine learning in trading: theory, models, practice and algo-trading - page 1665

 
Igor Makanu:

A vivid example of this is the MQL5 community application store. Behind the MQL5 brand stands a professional trading platform - MetaTrader5 and a huge community of traders and programmers. So, this store has a section for neural network solutions, and there are a lot of bots there, but none of the real ones. It's not allowed to add a bot that uses neural networks by the rules. I find that hard to believe, but it's true. The ban on external API connection disables neural networks and built-in MQL5 tools do not work in practice. I discussed this issue with MQL5 Technical Support, but received no clear answer.

This is either a hit on Metaquotes or the author's unprofessionalism.

It's more likely a primitive marketing attempt to promote a product that I cannot guess which category this product belongs to.

...Yes I remember! On the same hubra not long ago I read about some application that determines the nationality by photo/picture, there was one comment, they promised you an entertainment product, not a scientific program ))))

Hmm, how is it... Are you sick?

;)

And what in the quote is essentially wrong?
 
Evgeny Dyuka:
but what in the quote is wrong in essence?

it is not correct that the language does not support neural networks and something is forbiddenhttps://www.mql5.com/ru/search#!keyword=%D0%BD%D0%D0%B5%D0%B9%D1%80%D0%BE%D1%81%D0%B5%D1%82%D1%8C&module=mql5_module_market

by your statement you are questioning the work of the authors of the articleshttps://www.mql5.com/ru/search#!keyword=%D0%BD%D0%D0%B5%D0%B9%D1%80%D0%BE%D1%81%D0%B5%D1%82%D1%8C&module=mql5_module_articles

note that some of these products in the marketplace are posted by the authors of articles, the material of articles posted free of charge and many solutions from articles can be simply repeated and posted in the marketplace.

ok, it's up to you, but imho, you made an anti-publicity statement

 
Igor Makanu:

it is not right that the language does not support neural networks and something is forbidden

I looked through the list of products. The first one in the list by relevance has one comment and that's without text, and this is from 2014, has anything more relevant appeared over that time?
The article is also from 2014, a lot of water has flowed since then...

I think we should not be offended, but just hear the criticism and change faster. MQL5 crypto has already been lost and neural networks will also be missed.

 
Igor Makanu:

it is not right that the language does not support neural networks and something is forbidden

As for "forbidden", we have already discussed it, you just need to browse the forum above.
 

My robot died yesterday in agony ))

Evgeny Dyuka:

You have 20-30 networks working and I had 696 :)

 
mytarmailS:

My robot died yesterday in agony ))

You have 20-30 networks working and I had 696 :)

That's what I mean too, it's not going to last. Learn to identify the moment when it starts to drain. And yes, working with NS on the market, it's a training it 2 times a week.... I don't always get to train it, that's why I miss the moment of the plunge...
 
Mihail Marchukajtes:
I mean it will not last forever. You should learn to identify the moment when you start plummeting. And yes, working with NS on the market, it's a training it twice a week.... I don't always have time to train, that's why I miss the moment of drawdown...

The funny thing is that it died after I trained it with new data...

I also noticed that after training the robot does not always start trading normally right away, sometimes I have to skip a day or two of trading and then...

I'll illustrate...

somewhere

1) I totally retrained here

2) The robot made a little profit and then it emptied

But after a day or so it started working again.




Does your robot trade normally at once?

By the way, my approach is very similar to yours

 
mytarmailS:

The funny thing is that it died after I trained it with new data...

I also noticed that after training the robot does not always start to trade normally immediately, sometimes I have to skip a day or two of trading and then...

I'll illustrate...

somewhere

1) I totally retrained here

2) The robot made a little profit and then it emptied

3) But after a day or so it started to work again

You say, what are you talking about, it's just chaos, the robot is not working and simply retrained and not adequate to the market

I say maybe. But

1) I see exactly the same picture for not even the third time

2) it's hard to make a hundred deals a day(like last time) and be in the good profit if you re-trained or accidentally


Here is the same trade of the same losing robot but when he already started to earn, just for the last day

Look how clearly he sees the trends and opens in the right direction, can it be accidental?


See how he does today

 
mytarmailS:

You will say, what are you talking about, it's just chaos, the robot is not working and just retrained and not adequate to the market

I say maybe... But

1) I see exactly the same picture for not even the third time

2) it's hard to make a hundred deals a day(like last time) and be in the good profit if you re-trained or accidentally


Here is the same trade of the same losing robot but when he already started to earn, just for the last day

Look how clearly it sees the trends and opens in the right direction, can it be accidental?


EH... I recognize myself when I was young :-). Now teach the robot and run the test on the whole quote history. And you'll see that the balance curve will indeed go steadily down and then steadily up. That is, it may work or not, so you can go further and really make the second level of the neural network, but it will be difficult to trace the two. I.e. you train polynomial and make it a basic strategy, keeping an eye on how its parameters work and making predictions on the start of the set, the main thing here is to jump out of the set in time. That's one approach. The other approach is when you stably get a working TS after the training, even if it's not for a long time. The main indicator on the market is not profit, but the stability of profit, in the second case, you need to get 80% of profitable TS. That is, out of 10 weeks to lose a maximum of 2. But as practice shows, to lose in general is impossible. The only difference between these two is that the market is not really profitable. I was just moving in on myself.

In fact, you've got an approximation with the polynomial of your series, which can be used as a basic strategy, it's an indicator. Build it on the whole history and you'll be amazed :-)

 
Yeah, I guess I'm just dumb and wishful thinking.