Machine learning in trading: theory, models, practice and algo-trading - page 3275

[mytarmailS]

about the ansables of strategies

https://buildalpha.wordpress.com/2018/11/20/buildalpha-ensemble-strategies-reduce-overfitting-by-combining-strategies/

----------------------------------------

So all you need is a metric of strategy retraining to know whether a strategy will work on new data or not, everything else is solvable....

----------------------------------------

There is an idea to take several approaches to detect overtraining, mine is based on auto.arima, Prado "PBO". something else is possible, throw in as predictors and teach AMO to predict the probability of overtraining and make it a metric.

Alternatively.

[Maxim Kuznetsov]

mytarmailS #:
strategy retraining metric to know whether a strategy will work on new data or not.

another variant of the word grail ? :-) "to know if it will work in the future"

[mytarmailS]

Maxim Kuznetsov #:

another variant of the wording of the word grail ? :-) "to know if it will work in the future"

I am not precise, we need an honest probability for example - will work on new data with 69% probability.

[mytarmailS]

about p-hacking and strategies

https://mathinvestor.org/2019/04/p-hacking-and-backtest-overfitting/

[Lorarica]

If I may, I would like to contribute to such an interesting and important topic.

Machine learning, ML, is a class of artificial intelligence methods, the characteristic feature of which is not a direct solution to a problem, but learning by applying solutions to a set of similar problems. To construct such methods, tools of mathematical statistics, numerical methods, mathematical analysis, optimisation methods, probability theory, graph theory, and various techniques for working with data in digital form are used.

P.Z.

[lynxntech]

Lorarica mathematical statistics, numerical methods, mathematical analysis, optimisation methods, probability theory, graph theory, and various techniques for working with data in digital form are used.

P.Z.

the most important thing here is not to write a lot of words, already so tired of searching that 1-2 sentences will not be read

[Maxim Dmitrievsky]

mytarmailS #:

about p-hacking and strategies

https://mathinvestor.org/2019/04/p-hacking-and-backtest-overfitting/

So strategy optimisation or learning should look something like this:

where the mean score is the analogue of the result when using cross-validation

R2: 0.9849988744314404
Learn 1 model
R2: 0.9689143064621495
Learn 2 model
R2: 0.987424656181599
Learn 3 model
R2: 0.9439690206389704
Learn 4 model
R2: 0.9814487072270343
Learn 5 model
R2: 0.9636828703372952
Learn 6 model
R2: 0.986048862779979
Learn 7 model
R2: 0.960923469755229
Learn 8 model
R2: 0.9734744911894477
Learn 9 model
R2: 0.983760998020949
Learn 10 model
R2: 0.970035929265801
Learn 11 model
R2: 0.9888147318560191
Learn 12 model
R2: 0.9724422982608569
Learn 13 model
R2: 0.9554046278458146
Learn 14 model
R2: 0.9664401507673384
Learn 15 model
R2: 0.9806752105871513
Learn 16 model
R2: 0.977769556127485
Learn 17 model
R2: 0.9760342284284887
Learn 18 model
R2: 0.9769043647488534
Learn 19 model
R2: 0.9741849376008709
Learn 20 model
R2: 0.9740162061450146
Learn 21 model
R2: 0.919817531536493
Learn 22 model
R2: 0.9788269230776873
Learn 23 model
R2: 0.9579249703828974
Learn 24 model
R2: 0.9612684327278544

>>> o[0].mean()
0.9706082542553089

>>> o[0].std()
0.015284036641045055

[mytarmailS]

Lorarica #:

It is not a definition, not a complete set of properties.

A definition should clearly answer the question of what it is.

What you have is just a bunch of words from the internet with the tag MO.

[mytarmailS]

Maxim Dmitrievsky #:

So optimising a strategy or training should look something like this:

where the mean score is the analogue of the result when using cross-validation

Roughly like this... is this how?

What is o[0] ?

[Maxim Dmitrievsky]

mytarmailS #:
That's about right... how's that?

What is o[0] ?

Oh, everything.