Machine learning in trading: theory, models, practice and algo-trading - page 94

New comment
 
Mihail Marchukajtes:
Well??? Anyone managed to unravel my data????
I wish I had all the tools, I'm on vacation. I'll spin some more.
 
Alexey Burnakov:
It's a pity you answered early. You could have tortured the training methods.
Mihail Marchukajtes data spinning, but I feel it is there kukuveritel, d. gopher dig up))))
 
mytarmailS:
I don't understand how this "predictability" is calculated and whether it makes any sense if the target is not taken into account
My understanding from the article to this package is that it's the predictors themselves that are predicted, not the target variable.
 
Mihail Marchukajtes:
It's not about training methods. The use of lag in predicates will not allow to build a sufficient model. After all, a line is fed to the model for consideration, and if the data from the previous line is used, the model simply will not see it, so the task is obviously unsuccessful.
A neural network with memory could probably do the job. Plus genetics for selecting predictors. Genetics makes sets of predictors and selects neuronka parameters, neuronka is trained. Crossvalidation to check the quality of the model. Somehow it will work. Forex is more complicated, on it the lag dependence simply will not exist in the test file.
 
Vizard_:
Mihail Marchukajtes the data spin, but I feel he's got it wrong, you can't dig up a gopher))))
No, one multiplication and one division. Seriously....
 
Dr.Trader:
A neural network with memory would probably do the trick. Plus it would have genetics for selecting predictors. Genetics makes sets of predictors and selects neuronka parameters, neuronka is trained. Crossvalidation to check the quality of the model. Somehow it will work. Forex is more complicated, on it the dependence of lags simply will not exist in the test file.
Unfortunately, the network cannot cope with memory either, how can the network cope with data that it does not see, which is in another record that it has not yet been fed, and that other record must be fed exactly with the current record to catch this dependence...
 
SanSanych Fomenko:
My understanding from the article to this package is that the predictors themselves are predicted, not the target variable.

Well, yes, but damn, it's not right, a quality predictor is the one that explains the target well, not the one that explains itself, I do not understand how you can know the quality of the predictor without comparing it to the target, I do not understand....

what's in the archive?

 
mytarmailS:

Well, yes, but damn, it's not right, a quality predictor is the one that explains the target well, not the one that explains itself, I do not understand how you can know the quality of the predictor without comparing it to the target, I do not understand....

what's in the archive?

An article that explains it all
 
I will not torture the eesel formula, if you open a text file exactly in excel =C3*C4/F3. As you noticed you need to multiply the variable v2 by your lag and divide by v3 . And it does not work because the lag network does not see. Another thing is if you operate with data within one line, without using lag, you will get a model, no matter how sophisticated the formula is. By the way you should try to make up a mega formula within one line and see if the network can cope or not. I'll try it now....
 
Mihail Marchukajtes:
"Gopher" I could not find, screwed up))) And what are all these exercises for... Who "hippies" will understand.
1...87888990919293949596979899100101...3493
New comment
Reason: