ForumSections

Machine learning in trading: theory, models, practice and algo-trading - page 3182

New comment
fxsaber #:

for Monte Carlo is probably the best option for generating a random symbol with the desired statistical characteristics.

Lice test with the same set.

Top - real, bottom - random.

Failure.

The difference is that nothing is looked for in the optimisation results. That is, the first five optimisation results have never been run on OOS before.

Forum on trading, automated trading systems and testing trading strategies

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

Valeriy Yastremskiy, 2023.08.17 10:44 AM

It seems logical that the probability of false positives and negatives decreases, but to me it is not a panacea against errors.

Do you think you should trust the train_optim + test_forward model more than (train+test)_optim?

Aleksey Vyazmikin #:

There's no error. No error. No error.

I took another sample - 47% of units and examples 5 times less - after a couple of passes I can say that it finds many quantum segments already.

What else I paid attention to is the similarity of the initial target and the generated one - they are 49% similar - maybe it is necessary to set some threshold here - no more than 30% similarity? Any thoughts?

I made a graph - if there is a match "-1", if there is no match, then "+1", I got such a balance.

A lot of trends on it, what do you think?

fxsaber #:

Top is real, bottom is random.

I made optimisation with the same settings on the real and on the generated symbol.


Real.


Random.


It is not trained on Random.

[Deleted]
fxsaber #:

Did an optimisation with the same settings on the real and on the generated character.


Real.


Random.


Not trained on Random.

How many times was the random symbol generated?) obviously, the longer the sequence, the more times it needs to be generated to work on it too.
fxsaber #:

The difference is that nothing is searched for in the optimisation results. I.e. the first five optimisation results have never been run on OOS before.

I don't know how to put it simply.

You pick better variants "by hand" on OOS after optimisation on the test and it's NOT fitting...

And if the algorithm selects the best variants on OOS after optimisation, it's already fitting... Why?

Choosing the best options/variant from the full set of options is optimisation.... It doesn't matter if you do it by hand or by algorithm.

Perhaps you have only worked with a tester in MT and think a bit formulaic about optimisation itself and ways of its application, that's why we have some misunderstanding


fxsaber #:

Did an optimisation with the same settings on the real and on the generated character.


Real.


Random.


It's not trained on Random.

Does random match the price characteristics? Mean, standard deviation, covariance?

Maxim Dmitrievsky #:
Сколько раз был сгенерирован рандом символ?)

One.

it is obvious that the longer the sequence, the more times it needs to be generated to work on it too.

I don't understand this statement. What is meant by the following two options?

  1. You need to run many iterations of randomisation to get it to work.
  2. If you create a lot of randomised characters, the probability increases that there will be a workable one among them.

Seemingly simple things, because even on different real symbols the same TC does not work.

The algorithm of randomisation is as follows:

  1. A real tick history is taken.
  2. A sequence of increments of the average ((bid+ask)/2) price is made from it.
  3. In this sequence each term is randomly multiplied either by +1 or -1.
  4. A new tick history is collected from the obtained sequence of increments, where time and spread coincide with point 1.
  5. The new tick history is written in a custom symbol.
I.e. some real symbol is randomised. You can apply item 3 any number of times. If after item 5 all five points are repeated, it is the same as repeating item 3 twice.
[Deleted]

Exactly the same way I get both OOS working models and not, through the same algorithm. The symbol is the same, no new randomisation has been added. It just randomly finds either long-running patterns or local patterns. Because training on a random subsample (40% inside the line-selected range), and still stand other randomisers looking for patterns through randomisation, which randomisation chases.


1...317531763177317831793180318131823183318431853186318731883189...3749
New comment