Machine learning in trading: theory, models, practice and algo-trading - page 2054

New comment
 
Maxim Dmitrievsky:

What do you mean the correct result? These are errors for different datasets

it's not the time series that is sampled, but the labels. see the videos

what are the labels attached to? )))

just one question ! you sample and then split it into a track and a test ?

or vice versa ?

 
Maxim Dmitrievsky:

You can't use algorithms that take so long to learn... you can go gray.


Now I'll try to increase the coefficient, it should go faster, but there is a chance that I will have to re-learn

 
Alexander Alexeevich:

I understand correctly, that you teach the network to predict time series?

How much is enough )) it's not a network at all!)

 
Alexander Alexeyevich:

Akurashi as I understand it is prediction accuracy? And logloss? There should not be training on the test, and the error should be the same regardless of the number of passes? or -+ at least, but it should not decrease

Accuracy, logloss is a cross-entropy, all classifiers are trained by it. Then the accuracy is measured and displayed at each iteration for trayn and test

training only on a trayn, of course
 
mytarmailS:

What are the tags attached to? )))

just one question ! do you sample and then split into a track and a test ?

or vice versa ?

first sampling, then splitting

you can do it the other way around, the result will be the same since the sampling is random

i gave the function earlier, nobody reacted

 
Alexander Alexeyevich:

I understand correctly that you train the network to predict time series, right?

classification, signals 1-0

 
mytarmailS:

I've been here for a while))) Trying to reread the posts, but it's taking too long.)

I've been here not long))), tried to re-read the posts, but it's a long time))))) and a lot of rubbish, useful information in 5 percent of all posts)) although maybe lately much more

 
While you guys are flubbing, I need to parse the model and watch the bot in the MT tester
 
Alexander Alexeyevich:

Now I will try to increase the coefficient, it should go faster, but there is a chance of overtraining

overtraining will be in any case without an early stop

 
Have you ever tried time series prediction?
1...204720482049205020512052205320542055205620572058205920602061...3495
New comment
Reason: