Machine learning in trading: theory, models, practice and algo-trading - page 85

New comment
 
Mihail Marchukajtes:
Try cumulative delta. Cumulative distribution on real volumes...

Where do I get real volumes in the form of historical data? MetaTrader provides only the tick counter, which is called "volumes". Moreover, the values of these counters may differ by orders of magnitude in different kitchens.

Mihail Marchukajtes:
Data on other pairs, even on exotic ones that are not connected with the predicted one...
It will be necessary to try the indices and oscillations of other financial instruments that correlate with the analyzed one. At least, they can be found in MetaTrader in order not to depend on third-party sources of information.
 
Yury Reshetov:

Where do I get real volumes in the form of historical data? MetaTrader provides only the tick counter, which is called "volumes". Moreover, the values of these counters may differ by orders of magnitude in different kitchens.

It will be necessary to try the indices and oscilli of other financial instruments that correlate with the analyzed one. At least, they can be found in MetaTrader in order not to depend on some external information sources.
The real volumes of futures currencies such as the euro, the pound, the yen. There is also the concept of delta, which is the number of buyers and sellers at a given moment and at a specific price....
 
The market is two-faced in every sense. As yin and yang.... That's why it's difficult to imagine a system which trades long and steadily, because the market is a living organism. So, paying attention to the delta, I will say. Sometimes the market goes with the crowd, sometimes against the crowd. As in the pound today. Sellers are more and the market goes up, buyers are more and the market goes down - this is called a bait for the crowd, but in other periods the market goes with the crowd, there are many buyers and the market grows. I think maybe the classifier should work in this direction. To determine the strength or weakness of sellers or buyers...... Then the accuracy of market conditions will be a system with minimal risk and a high degree of profitability..... Only so far I do not know what to do.....
 
Alexey Burnakov:
To read:https://www.researchgate.net/publication/236896558_FOREX_Daily_Trend_Prediction_using_Machine_Learning_Techniques

You can read such "scientific" articles at least until you lose your pulse. After all, such scientific crap in the Internet a carload and a small cart, even in the Russian-speaking segment. Someone needs to get another academic degree or to increase the number of "scientific" publications, so they compose such disreputable crap. However, you can't put this kind of reading on your bread and put it in your pocket. Since the text of the publication can neither prove nor deny the phrase that supposedly "Four important Forex currency pairs are investigated and the results show consistent success in the daily prediction and in the expected profit". There are just a lot of figures, graphs, smart mugs on the authors' avatars and no details about monetization.

Normal authors would have given a link to the original data: the samples for classification and methods of obtaining them. But in this case, as in most other pseudoscientific articles, the authors are afraid to do so, because then anyone who has mastered some binary classifier, can easily double-check all this stuff for "lousiness" and catch their hand in the fact that the market stability can be dreamed, not ranted.

 
Yury Reshetov:

Reading such "scientific" articles is possible at least until the pulse stops. After all, such scientific crap in the Internet a carload and a small cart, even in the Russian-speaking segment. Someone needs to get another academic degree or to increase the number of "scientific" publications, so they compose such disreputable crap. However, you can't put this kind of reading on your bread and put it in your pocket. Since the text of the publication can neither prove nor deny the phrase that supposedly "Four important Forex currency pairs are investigated and the results show consistent success in the daily prediction and in the expected profit". The article contains just a bunch of figures, some charts, smart faces on the avatars of the authors and no details about monetization.

Normal authors would have given a link to the raw data: samples for classification and methods of obtaining them. But in this case, as in most other pseudoscientific articles, the authors are afraid to do so, because then anyone who has mastered some binary classifier, can easily recheck all this stuff for "lousiness" and catch their hand in the fact that one can dream about market stability, not rant.

Well, I agree with that.
 
Yury Reshetov:

Reading such "scientific" articles is possible at least until the pulse stops. After all, such scientific crap in the Internet a carload and a small cart, even in the Russian-speaking segment. Someone needs to get another academic degree or to increase the number of "scientific" publications, so they compose such disreputable crap. However, you can't put this kind of reading on your bread and put it in your pocket. Since the text of the publication can neither prove nor deny the phrase that supposedly "Four important Forex currency pairs are investigated and the results show consistent success in the daily prediction and in the expected profit". The article contains just a bunch of figures, some charts, smart faces on the avatars of the authors and no details about monetization.

Normal authors would have given a link to the raw data: samples for classification and methods of obtaining them. But in this case, as in most other pseudoscientific articles, the authors are afraid to do so, because then anyone who has mastered some binary classifier, can easily recheck all this stuff for "lousiness" and catch their hand in the fact that the market stability is something to dream about, not to rant.

Yury say.... In old versions of HSPF divided the sample in half, but now it is skewed, as far as I understood. What is this associated with, and is it possible to return the division in half in the current version of the binary-trend classifier. There are good reasons for this......Now it is difficult to explain, but if necessary I will think up and give examples.... There is simply a possibility to return the sampling splitting in half??? Maybe put a checkmark. if there is a checkmark, divide in half, no, divide with an offset....
 
Mihail Marchukajtes:
Yura say.... In older versions of HSPC divided the sample in half, but now it is skewed as far as I understand.

It is balanced in the training part (the number of examples for both classes is the same) and not necessarily balanced in the test part. If splitting it in half, then we can only hope for luck in terms of balancing.

The HGCP does not split the sample, but mixes the examples in it with an even distribution before splitting it.

 
Yury Reshetov:

It is balanced in the tutorial part (the number of examples for both classes is the same) and not necessarily balanced in the test part. If you divide it in half, you can only hope for luck in terms of balancing.

The PRSG doesn't split the sample, it shuffles the examples in it with a uniform distribution before splitting it.

Well let's say he shuffled and split in half, it turns out that the training and test samples will have the same number of both classes, right?

 
Here's the thing, let me try to give you an example...
1...787980818283848586878889909192...3522
New comment
Reason: