Machine learning in trading: theory, models, practice and algo-trading - page 59

New comment
 
Regarding methods, I suggest you read the article. The industry has an empirically formed view of which methods are best. This article is about the classification problem : http://www.cs.cornell.edu/~caruana/ctp/ct.papers/caruana.icml06.pdf

And there is no need to invent anything.

But there are exceptions, of course. It all depends on the data.
 
Dr.Trader:

In other words "Look at the first google link that leads to my site" :)

No one forbids you to look at other links. I did not put a link to my site at the top of google, did I? So all complaints about the fact that google indexes the site is not as you would like to, do not address to me, and to a support team of a search engine.

Dr.Trader:

I found, you have a committee of two models, this is not how I understood and wrote above.

I'm sorry, but I'm not trained in a telepathic course to read from a distance what you meant. So either address your complaints to the Ministry of Education as to why telepathic training is not included in the mandatory curriculum for all educational institutions. Or be more specific in your thoughts on this form.
 
Mihail Marchukajtes:
I rarely had models rise above 40-50% generalization, but after I thought about what to do with the data. What is the essence of the model obtained after the classification. On the same data I now get models no lower than 70% on average 80-90% and in the future, on unknown data errors of about 1-2 out of 10-12. It is quite enough to earn.)
Maybe you can tell me what kind of miracle transformation you should do to raise the quality of recognition
 
Dr.Trader:
Yeah, I'm trying to classify strictly buy/sell too. But how did you get the original 6 inputs, did you just take them from some known strategy? Adequate entries is one of the most important things. On the contrary, I have thousands of entries (prices and indicators over a hundred bars) and I need to sift them out leaving a couple dozen, because any model overtrains on so many inputs.

My strategy is simple. This is Thomas Demark's sequent, which gives buy and sell signals. Signals that are more profitable than 100 pips are marked with one, the rest are all zeros and at the time of the signal I save the values of the indicators set below and get a model of about 90% generalization. That's all...

You can also take the crossing of the bars as a basis for the system. I also think it should be good. So that's how it is. The main thing is to prepare the data correctly...

The last two are zetoscore models and kelly coefficient, so nothing extravagant....

  double PONT1=iBullsPower(NULL,0,PONT,PRICE_CLOSE,i)+iBearsPower(NULL,0,PONT,PRICE_CLOSE,i);
  double MOM=iMomentum(NULL,0,PONT,PRICE_CLOSE,i);
  double Dstoh=iStochastic(NULL,0,PONT,PONT+9,PONT+9,MODE_EMA,0,MODE_MAIN,i);
  double STD=iStdDev(NULL,0,PONT,0,MODE_EMA,PRICE_CLOSE,i);
  double Force=iForce(NULL,0,PONT,MODE_EMA,PRICE_CLOSE,i);
  double VolM=iCustom(NULL, 0, "BetterVolume 1.4.1",9,i);
  double EMA=Close[i]-iCustom(NULL, 0, "9MAMA_NK",0.5,0.05,1,i);
  double Zscore=iCustom(NULL, 0, "TDSEQUENTA v2015",5,8,12,9,i);
  double Nstep=iCustom(NULL, 0, "TDSEQUENTA v2015",5,8,12,10,i);
 
After the red line of Out of Samples or Out of Samples for Beginners. I think it's quite viable. True, there have been some mistakes today, but that's okay..... There's no such thing as a mistake....
 
Mihail Marchukajtes:

My strategy is simple. This is Thomas Demark's sequent, which gives buy and sell signals. Signals that are more profitable than 100 pips are marked with one, the rest are all zeros and at the time of the signal I save the values of the indicators set below and get a model of about 90% generalization. That's all...

I.e. classification by future profitability (1 - not less than 100 points, 0 - less than 100 points), but not by signal direction? And how do you determine the direction, by the Demark's Sequence?

 
Yury Reshetov:
I.e. by the profitability classification (1 - at least 100 pips, 0 - less than 100 pips), but not by signal direction? And how do you determine the direction, by the Demark's Sequence?

The system itself gives a buy signal or sell signal, this is the direction, and the classifier already says if the signal is buying and NS says yes this is the right signal, then buy, if it says no this is not the right signal, then sell. It is the same with selling.... True or false sale, hence the conclusions ...

 
Mihail Marchukajtes:

The system itself gives a buy signal or sell signal, this is the direction, and the classifier already says if the signal is buying and NS says yes this is the right signal, then buy, if it says no this is not the right signal, then sell. It is the same with selling.... True sale or false sale, hence the conclusions...

Can you place an example of a sample for classification in CVS on the forum?
 

And it is possible to bring any input data to the output data and the system will work for some time, so he who is looking will always find :-)

And so, indeed, a small prank with the data and we have a level of generalization grows to acceptable numbers in 90%.....

 
Mihail Marchukajtes:

And it is possible to bring any input data to the output data and the system will work for some time, so he who seeks will always find :-)

So, really, a little trick with data and we have a level of generalization grows to acceptable numbers in 90%.....

What is the duration of the data for training and validation? It looks like a couple of days? It doesn't say anything at all, to be honest.
1...525354555657585960616263646566...3493
New comment
Reason: