Machine learning in trading: theory, models, practice and algo-trading - page 2751
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
Now that's a really big step forward. Do indicators and experts need to be rewritten?
while you are here, thanks for ICA and still a question about feature_extraction(which you characterise as"recognition of unique and significant components") -- I was browsing this topic once (description for Python), and then abandoned it.... I got the impression that they use this extraction only when working with images and text -- should we bother with this approach when working with structured data (like a table of factors) ??... can you write a few words about the essence of the algorithm to understand its usefulness?, just in your article somehow glimpsed about it...? or did I miss something? (and what I saw in Python didn't inspire me to apply feature_extraction to factors in the market).
on preferences: or did you still decide to stay with regular (or hierarchical, as you described) PCA ???
what dimensionality_reduction are you using now ?
and thanks for the article
there are significant doubts that NNs can work at all with increasing information. Just above there was a reference to something running between the walls and eating green without touching red. As the maze increases (information growth) it will have its memory of red/green blown away. So he moves well in the previous formation, but not in the new one. As long as the patterns are repeated, he works more or less. And it works perfectly only where it is trained.
The counter-thesis to the same NN for the same conditions is that a student can write a more efficient algorithm. And this algorithm another student can improve or bring to conditions of other_curvature_space(or optimise/simplify). Using NN in this simplified example is a dead end, its result cannot be consciously and qualitatively improved. And this is generally a general problem with NN (and DL too). These methods can be used to recognise red/green, round/square from a set of reddish, greenish, on a background of purple, in B/W oblong. The solution of narrow special problems, not the output of equity to infinity
KNN is not the same thing at all.
I am interested in "predictive ability", not classification, and even without a teacher, which are useless in our business.
Nevertheless, KNN gives the easiest way to deal with non-stationarity if you use it in a sliding window. It uses the same set of predictors all the time, which makes it possible to compare it with an algorithm that changes the set of predictors and evaluate the benefit of this complication.
I will post my method, which solves similar problems robustly and sensitively - I will give the theory and post the code in R.
This is done for mutual enrichment of "understanding" of machine learning tasks.
Is what I have posted not enough to post what you have said?
However, KNN provides the easiest way to deal with non-stationarity when used in a sliding window. It uses the same set of predictors all the time, which makes it possible to compare it with an algorithm that changes the set of predictors and evaluate the benefit of this complication.
I don't see the point in doing something that can't produce the desired result.
Who here asked about application of NN and AI in trading ?
(moderated) platform (moderated ) ( moderated) (moderated) (moderated) (moderated) (moderated) understands the description of trading conditions/rules/parts-algorithms in natural language. Of course in English
Here in my opinion is the right direction of movement and application of AI efforts. As in its Wolfram domain, but Wolfram is mostly an encyclopaedia
---
right now, people who are in the same field and have been communicating for a long time are not able to understand each other. Without an auto-translator or a mutually signed and binding agreement on naming entities:-)
Have you done a comparison of your algorithm with KNN (or some modification of it)? It would be interesting how significant the gain is.
I don't see the point in doing something that in principle can't give the desired result.
Nevertheless, KNN gives the easiest way to deal with non-stationarity if you use it in a sliding window. It uses the same set of predictors all the time, which makes it possible to compare it with an algorithm that changes the set of predictors and evaluate the benefit of this complication.
I can't agree with that.
The market is changing, and the time intervals of change are different and independent of each other.
I used to be able to write Expert Advisors that lived from 3 to 6 months. I optimised them at weekends. Then they died, and for a short time quite enough to drain the deposit. Not enough time to optimise. In the end the situation was even worse: after some time it turned out that there was a boundary, moving away from which it was impossible to select parameters.
There are longer periods of market changes: 5-7 years. But the result is the same as for monthly periods. The bot dies forever. I will send a specific bot from the market in a private message - you can't do it here.
So this whole idea of "out of sample" is rubbish. A bot still has a lifetime, we don't know how long: 3 months or 7 years. When the bot dies, we confuse it with another drawdown and drain our depo.
Ideally, it should be retrained on the next candle. If we work on ticks, then on the next tick, on H1 then on the arrival of the next hour.
Thank you, it is quite clear, why on every bar))))) And why after minutes you go to ticks))))))
if KNN is essentially K-means (foreign name for k-nearest-neighbors?), then it(K-means) uses Euclidean distance ... "when the variables are not correlated -- the Mahalanobis distance coincides with the ordinary Euclidean distance" .... in LDA . .. well if they are correlated -- then Mahalanobis is better, as noted yesterday in a general wayNothing prevents you from using any other distance instead of Euclidean distance. You can't use Mahalanobis, of course, since it's the distance between a point and a sample, not between two points.
I don't see the point in doing something that can't produce the desired result.
Just in case, I will say that I am not going to tell you what you should do. Rather, I am just thinking aloud what I would do myself in case there is an algorithm that gives good results but is poorly implemented in the form of an Expert Advisor (although I always try to avoid this option). Most likely, I would try to get similar results using algorithms that are easier to implement. Among other things, I would get some analysis of what exactly the algorithm is good at.
I would start with KNN and if it gives a similar result, then it is a matter of good selection of a common set of predictors. If the result is much worse, then maybe it's just a matter of choosing a subset of predictors at each time point. To test this hypothesis, I would try to use local regression (something like LOESS), since regression already allows you to compare the significance of predictors. Further steps are already based on the results of the analysis. By the way, with the appearance of matrices in mql5, linear regression has become easy to implement directly in it.