Machine learning in trading: theory, models, practice and algo-trading - page 3380
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
In MO, the fitness function is used to train the model (parameter selection) through optimisation. A metric(s) is used to evaluate the resulting model. Often the metric does NOT match the fitness function. From a mathematical point of view, this means that the MO is solving a MULTICRITERIAL optimisation problem rather than the usual single-criteria one.
Another significant difference from conventional optimisation is the frequent absence of a fixed set of optimisation prameters. Even for a regular tree this is already the case. From a mathematical point of view, this leads to an optimisation problem in FUNCTIONAL SPACE instead of the usual one in numerical space.
Both of these points make MO problems irreducible to conventional optimisation.
There is a whole separate class of multi-criteria optimisation algorithms. But, if properly understood, multicriteria is reduced to additional boundary conditions and separate evaluations.
Functional space also requires evaluation. Everything always requires evaluation.
In MO, the fitness function is used to train the model (parameter selection) through optimisation. A metric(s) is used to evaluate the resulting model. Often the metric does NOT match the fitness function. From a mathematical point of view, this means that the MO is solving a MULTICRITERIAL optimisation problem rather than the usual single-criteria one.
Another significant difference from conventional optimisation is the frequent absence of a fixed set of optimisation prameters. Even for a regular tree this is already the case. From a mathematical point of view, this leads to an optimisation problem in FUNCTIONAL SPACE instead of the usual one in numerical space.
Both of these points make MO problems irreducible to conventional optimisation.
Thanks for the detailed explanation. Still, there was some context for raising the topic of FF. Here it is.
Forum on trading, automated trading systems and testing trading strategies
Machine learning in trading: theory, models, practice and algo-trading
Maxim Dmitrievsky, 2024.01.10 19:27
Well through re-optimisation with a check on OOS can be found :) which is the simplest case of a wolf forward OR cross validation with one fold.There is a whole separate class of multi-criteria optimisation algorithms. But, with proper understanding, multicriteria is reduced to additional boundary conditions and separate evaluations.
Functional space also requires evaluation. Everything always requires evaluation.
The features I mentioned work simultaneously, not one by one, so I don't know what kind of boundaries you are going to build in functional spaces.
It would be more useful if all participants of the thread were familiar with the basics of modern MO. A textbook from SHAD would be a good option.
The features I mentioned work not one by one, but simultaneously, so I don't know what kind of boundaries you're going to build in functional spaces.
Yes, we are talking about simultaneous work of separate components in multifunctional spaces. Both components can be evaluated separately in a multifunctional space and all together - by meta-evaluations, or otherwise - by integral evaluations. One does not interfere with the other. Any stages of MO require evaluations, for this purpose there are many special metrics, maximisation of which is the essence of optimisation.
1) In MO, the fitness function is used to train the model(parameter selection) through optimisation. A metric( s) is used to evaluate the resulting model. Often the metric does NOT match the fitness function. From a mathematical point of view, this means that the MO is solving a MULTICRITERIAL optimisation problem rather than the usual single-criteria one.
2) Another significant difference from conventional optimisation is the frequent absence of a fixed set of optimisation prameters. Even for a regular tree this is already the case. From a mathematical point of view, this leads to an optimisation problem in FUNCTIONAL SPACE instead of the usual one in numerical space.
Both of these points make MO problems irreducible to conventional optimisation.
1)
What is the contradiction ?
parameter selection == parameter search in the optimisation algorithm
model metric estimation == FF with akurasi estimation for example.
What do you disagree with here ?
2)
Can you elaborate on what you see as the problem? For example, I don't see
Thanks for the detailed explanation. There was still some context to the raising of the FF topic. Here it is.
I saw your question, but I can't say anything intelligible about it.
And I'm not a very good translator of Maxim's language)
Yes, we are talking about simultaneous work of individual components in multifunctional spaces. Both components can be evaluated separately in a multifunctional space and all together - by meta-evaluations, or otherwise - by integral evaluations. One does not interfere with the other. Any stages of MO require evaluations, for this purpose there are many special metrics, maximisation of which is the essence of optimisation.
Saw your question, but can't say anything intelligible about it.
And I'm not a very good translator of Maxim's language).
It's not about translation.
Forum on trading, automated trading systems and testing trading strategies
Machine learning in trading: theory, models, practice and algo-trading
fxsaber, 2024.01.10 19:43
Let's assume that 100 steps are made - we got 100 sets of inputs. If we form the average set according to the principle"each input is equal to the average of the corresponding input 100 sets", it is unlikely that this set will pass well the entire initial interval.
Forum on trading, automated trading systems and testing trading strategies
Machine learning in trading: theory, models, practice and algo-trading
Maxim Dmitrievsky, 2024.01.10 19:46
If not, there are no good sets at all, logically. In terms of confidence in the future.It is clear that a hundred sets depend on FF.
Provide references, if not difficult (articles, books).
I kept several hundred books on neural networks, MO, optimisation, mathematics in the archive. I gave a link to the archive. The archive was available to everyone in the cloud for several years, at the moment I do not support this archive, the cloud with the archive does not exist now.
A.P. Karpenko has many books on these topics, good books are by Simon D.
It's not about translation.
It's clear that the hundred sets are FF dependent.