Machine learning in trading: theory, models, practice and algo-trading - page 1598

New comment
 
Boris:

Communication is the greatest value!


It has been noticed that some pairs already show a "slide", i.e. the presence of a local maximum, after which the balance curve begins to look down

I would consider it as someone who has already started to manage this process, and sooner or later all the other pairs will meet the same fate

in this regard, it is better to look for solutions that do not have a "slide", but do not put them out to the public

Either that, or cluster on modes and see how they will switch on new data. For example, change purchases in time to sales at certain hours or something like that.

Usually modes last a long time, the main thing is to switch in time

 
Maxim Dmitrievsky:

Either that, or cluster on modes and see how they will switch on new data. For example, in time to change purchases to sales at certain hours or something like that.

Usually the modes last a long time, the main thing is to switch in time

I usually test all results on "buy/sell" changes

it usually doesn't work.

I need to think if it is possible to apply it on synthetics and go there

There is nothing to catch on such a chart


 

Seriously, just lucky.... After all, all MO is either lucky or not. Only the distance is different for everyone.

This one is almost over :-(

 
Maxim Dmitrievsky:

Either that, or cluster on modes and see how they will switch on new data. For example, in time to change purchases to sales at certain hours or something like that.

Usually the modes last a long time, the main thing is to switch in time.

that's not gonna work for months.

And in weeks there is too little data and it's hard to know how to sort them out.

Weeks have no Fridays or Tuesdays ))

 

Hello ! I am sure that the future is in neural networks. An example of this account is ...

But I have another idea about neural networks ...

Let's write an EA based on convolutional neural networks.

In Python with Keras and Tensorflow 2 libraries.

We upload screenshots of charts and let the network make a forecast based on past screenshots if the price goes up or down!

I'm not a programmer unfortunately I would have done it myself, I have to try it for the sake of interest ...

 
mtyvnel:

Hello ! I am sure that the future is in neural networks . An example of this account is ...

But I have another idea about neural networks ...

Let's write an EA based on convolutional neural networks.

In Python with Keras and Tensorflow 2 libraries.

We upload screenshots of charts and let the network make a forecast based on past screenshots if the price goes up or down!

I'm not a programmer unfortunately I would have done it myself, I have to try it for fun ...

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0212320

Forecasting stock prices with a feature fusion LSTM-CNN model using different representations of the same data
Forecasting stock prices with a feature fusion LSTM-CNN model using different representations of the same data
  • journals.plos.org
Forecasting stock prices plays an important role in setting a trading strategy or determining the appropriate timing for buying or selling a stock. We propose a model, called the feature fusion long short-term memory-convolutional neural network (LSTM-CNN) model, that combines features learned from different representations of the same data...
 
Boris:

this approach will not work for months

And in weeks there is too little data and it is not clear how to sort them

weeks do not have Fridays or Tuesdays ))

It is clear, if you want to.

 
Maxim Dmitrievsky:

Everything is clear, if you want it to be.

There are nuances on watch and so on, connected with dates transition, i.e. at 21-22 o'clock on London, because at the moment of dates transition the value of opener seems to allow to get something interesting, but as we know, at this moment spread can be widened "to the limit" and this can prevent from realizing this interestingness in practice
 

and here there is nothing to catch, as it seems to me

However it is possible to see some "stationarity" of the process after the "shelf" ))))

 
A little bit about terver or random wandering or whatever you like. We take an unfamiliar fortune teller, put her eye in front of her. We ask a very specific question that interests us - what will happen in the future. There are 140 cards (or so) in the deck. The fortuneteller takes out 3 cards that fully describe the details of the current situation, of which she knew nothing, and says what will happen in the future (this cannot be verified yet). We observe the further work of the fortuneteller with other also strangers, they too remain delighted. Everything is transparent, people are anonymous, no one knows anything about anyone, including the fortune teller, the session is free. After 5-7 observations, my brain broke and I left. Is the sample representative enough?
1...159115921593159415951596159715981599160016011602160316041605...3495
New comment
Reason: