Machine learning in trading: theory, models, practice and algo-trading - page 3214
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
You're not making yourself clear. train and test certainly don't overlap.
Here is a computational process (it is called training, optimisation - it doesn't matter). Are both samples involved in this process?
The classical way is like this.
there is
train
test
validation
The algorithm learns from train and sees only train,
but the trainer also immediately looks at the test, so that the results of train and test do not diverge,
It may also decide to stop the training on the basis that the results of the test train have started to diverge.
so train sees the algorithm when training
test sees the person during training but does not see the algorithm
validation sees no one until all stages of training are completed.
but this is classic, nowadays many algorithms combine train and test training.
What's strange is that 10 min ago you claimed that you don't know professionals, and now suddenly you do )
I can't call algo-traders, whose cornerstone of making money is patterns, whose future is doubtful, professionals.
If they have any respect for themselves, it is not for research, but for high-quality technical implementation. Nobody respects their trading idea. Everyone understands that at any moment a breakdown may occur and only skills/experience will be left.
I have not seen any quants among algotraders.
You don't understand the code?
train is a file on which the model is trained, i.e. a list of patterns is formed, if in a random forest, then about 100 patterns.
test is a file where the algorithm predicts the value of the target variable based on the patterns.
At the moment when the TS model was obtained, was test involved in the calculations?
classically
there is
train
test
validation
The algorithm learns from train and sees only train,
but the trainer also looks at the test, so that the results of train and test do not diverge,
It may also decide to stop learning because the results of the test traine started to diverge.
so the trainer sees the algorithm during training
test sees the person in training but does not see the algorithm
no one sees the algorithm until all stages of training are complete
but this is classical, now many algorithms combine training and test .
Thank you. Then OOS is validation. Not test.
Thank you. Then OOS is validation. Not test.
he spells it wrong. OOS - test, validation - the second subsample (along with traine) to evaluate (validate) the model.
The validation one can be equal to the test one or separate.
This separation came about because IOs often use the second subsample to stop training early. You could call it fitting to it, in a sense.
Therefore, they use 3 subsamples, one of which is not involved in training at all.I can't call algo-traders, whose cornerstone of earning money is patterns, the future of which is doubtful, professionals.
Is there any other way to earn systematically without exploiting patterns?
because making money systematically == patterns .
Aregularity is a regular stable relationship in quantities, properties and phenomena of objects. In mathematical law..... blah blah blah blah.
If a trader can trade patterns that die, but he can systematically find new ones, then we can say that the trader is trading patterns, because he has learnt to act systematically for his own benefit, he is trading patterns.
At the time the TC model was obtained, was the test involved in the calculations?
No. That's a strange question
Is there any other way to systematically earn money without exploiting patterns?
Martin.
Aregularity is a regular stable relationship in quantities, properties and phenomena of objects. In mathematical law..... blah blah blah blah.
Trading a pattern. It has not died down for a long time, but it does not let us relax. For example, if now to return the market that was a couple of years ago, the output would be much better. But there were no tools then that are available now. It's a kind of "offensive" and "defensive" arms struggle.
I know multi-millionaires in crypto: they invested their stray trading million in crypto when they were buying pizza for bitcoins. There are no such stories.
If a trader can trade patterns that are dying, but he can systematically find new ones, then we can say that the trader is trading patterns, because he has learnt to act systematically for his own benefit, he is trading patterns.
This is a philosophy.
No. That's an odd question
You don't do random sampling, just take the left part as train and the right part as test. The peeking will go away immediately.
Martin.
And if there is a strong recoil-free trend, will the martin work? And why won't it work?
It's a philosophy.
1)if an algorithm makes money on the market for a long time, it is a regularity, right?
2)and if you superimpose another algorithm on top of this algorithm, which will observe the first algorithm and collect statistics on it and give trading signals and earn money.
Then it's already a philosophy, not a pattern, according to you.
That's exactly what I described with the example of a trader...