Machine learning in trading: theory, models, practice and algo-trading - page 1753
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
local regressions, increments. It selects attributes by itself, leaving 10-15 out of 250
Ah, the numbers of hours through the van hottlocal regressions, increments. It selects attributes by itself, leaving 10-15 out of 250
ah, more clock numbers via van hottWhat are local regressions? (forest on regressions?)
local regressions are what? (forest on regressions?)
No sneak peek, fix lot
Econometrics + Forest :)
Why are you testing through the ass. Is it possible to train a model in the future to trade the present ?
Why do you have a trainee after the test? You should not do it that way!
Why are you testing over the ass. Is it possible to train a model in the future to trade the present?
Why do you have a trainee after the test? You can't do that!
Don't pick on the little things.
Don't pick on the little things.
It's not a little thing, Max! I've done that more than once, you can't do that! Make it normal and let's see!
It's not a small thing Max! I've done it more than once, do it right and we'll see.
I will have the same, but from left to right.
I don't like these cp's on bars, not enough trades
it will be the same, but from left to right.
Yes, not a fact! This is the first, and secondly, why do so, if you can do it normally, yourself fucked ...
I even had such a case - counted the target for all data, then divided into a test and a train and it turned out the target was with memory and the test data showed itself very well, and when I counted it again only on the test data was all sad, and this is generally implicit, light case, and you do consciously bullshit that is not feasible in reality
i don't like these ts on bars, few trades
i don't like them, they have few trades and reality has little in common either, every candlestick close is a time subjectivity, the data should be first transformed in some way
I even had such a thing - I counted the target for all the data, then divided it into test and triple track, and it turned out that the target was in memory.
I even had such a case - I calculated the target on all the data, then divided it into a test and a trace and it turned out the target was with memory and performed very well on the test data, and when I recalculated it only on the test data it turned out very sad, and it was an implicit, light case, and you consciously do things that are not realizable in reality
the deals are few and the reality has little in common, every candlestick close is a matter of time, the data must first be somehow transformed
don't be nervous ))
in the MT5 tester no peeping is possible
will be the same, but from left to right
I do not like these ts on the bars, not enough deals.
I have an interesting idea, a kind of useful chelege, but I don't like those from left to right.
I have an interesting idea, such a kind of useful journal
What if we create a dataset (one for all) with target and prices + various useful indicators and post it here, make a test and a trace, and then "test2" for full OOS check of the already trained model.
People will upload dataset and try to improve the quality of classification, if something works, it will be added to the dataset as a chip/indicator.
The result:
1) dataset gets better, the chips in it get better.
2) Models are improved
3) Understanding of how to work with features and AMO evolves.
4) Even though it is remote, it is a team work
5) Benefits for all, everyone will finally have a community, not a quarrel, and Challenge is an incentive to do better than anyone else.