Where is the line between fitting and actual patterns? - page 63
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
...
What do you think? - How do you like it?
The only negative side effect is that you have to optimize it twice. What do you expect... beauty needs money.
the idea deserves further development. there might be something in it.
at least for this particular cp, such an approach should increase "oopacity".
Yeah. It's a good one. We should try it out. )
The implementation is not clear at all. And the criterion is a necessary and sufficient condition. It's all a bit vague...
Then, by means of multicriteria analysis we make up the importance of statistical indicators, such as pf, mo, col.sd, etc. so that our test samples from among those that have passed the OOS are lined up in order from the top.
Next, voila! We have a holistic criterion, or rather a multi-criteria criterion (sorry for the taffatalogy) that allows us to optimize our multi-starred TS in such a way that GA would result in those samples that are most viable in feedback...
Oh)
So, Andrey, it wasn't all so useless?
Oh, how)
So it wasn't all for naught, was it, Andrew?
Of course it was. You have to get hit on the forehead with a rake (it often happens more than once), so that after your brain has been shaken up, new, actually promising ideas will be born...
...then it usually turns out that it's just a new rake beyond recognition...
so that once the brain has been shaken back into place, new, actually promising ideas will be born...
...then it usually turns out to be just a new rake unrecognizable...
yeah.
promising ideas are tested by a new rake :)
For a potentially robust system, it seems possible to select a coherent "complex" criterion, by which all passed CBs will line up without gaps, if they are sorted by applying this criterion. This is not possible for non-Robust ones - there will be gaps in the rows. This is actually a measure of robustness - the integrity of the list of passed CBs...
Good, you should firstly make full enumeration of all parameters without GA. Use maximal possible step (of course, we must not be too rough with step, otherwise...). This gives us the following:
1. Independence of the list from any criterion, as it is search of all parameters, but not optimization
2. making "true" criterion for further optimizations, but with GA using this criterion.
I.e. the technology is the following: total search->creation of criterion->use of criterion in subsequent optimization with GA.
so on.
A facet can be calculated by comparing results between Sample and OOS
Many neural network packages have a very convenient way of separating the sample into a training sample and a test sample. The boundary is very easy to identify. If there is no improvement at all on the test sample then it is a naked fit. I.e. the inputs are non-kosher and the TS can be discarded. In other cases, we look to the moment when the results continue to improve on the training sample, and no longer improve on the test sample - this is the very edge.
Another thing is that division of history data into test and optimized samples is not implemented in the MT* tester and therefore it is practically impossible to detect a border in the optimization process regardless of whether you discuss it or not.
Experimenting with TS operating with probabilities I noticed that the fit appears when some historical events have zero probability. Any event that took place in history must have a probability higher than 0.
OOC- It's a screw-up, self-defeating and the same kind of fitting. More precisely, one of the variants of survivor error.
There is no point in dividing the story into parts, as you will still end up selecting parameters that pass the OOS. Which amounts to the same kind of fitting. You pick a random good one out of a pile of others. You can run the whole story at once - the result will be similar.
OOS- It's a screw-up, self-defeating and the same kind of tinkering. More precisely, one of the variants of survivor error.
There's no point in splitting the story up, because you'll still end up choosing parameters that pass the OOS. Which amounts to the same kind of fitting. You pick a random good option from a pile of others. You can run the whole story at once - the result will be similar.
It's like this.
There is one more thing that may make you think when writing TC.
It goes like this.
There is another point that might make you think twice when writing the TS.
Ticks are chaos, not worth wasting your time. There is no pattern to them and there can't be. Regularities can be found in time frames comparable to economic cycles of 250-350 days or more. Moreover, this is the period of a pattern, and the patterns themselves cover an even longer period of time. And what seems to many people are really seeming regularities that have nothing to do with a real regularity.
Tiki is chaos, not worth wasting your time. There is no pattern to them and there cannot be any. Regularities can be found in time frames comparable to economic cycles of 250-350 days or more. Moreover, this is the period of a pattern, and the patterns themselves cover an even longer period of time. And what seems to many is really a seeming pattern that has nothing to do with a real pattern.
Yusufhoja, tell that to those cranks who buy seats for their equipment right in the server racks of exchanges to speed up access to exchange traffic (tick-flow) - for they are apparently idiots and do not know what they are doing!
Tiki is chaos, not worth wasting your time. There is no pattern to them and there cannot be any. Regularities can be found in time frames comparable to economic cycles of 250-350 days or more. Moreover, this is the period of a pattern, and the patterns themselves cover an even longer period of time. And what seems to many people are really seeming regularities that have nothing in common with real regularities.