Machine learning in trading: theory, models, practice and algo-trading - page 2128
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
Apparently I did not understand the question.
There is no model interpreter on MT5 with categorization predictors, and CatBoost with command line can do everything on the idea that python version can do, except pure python things, such as visualization.
Did you make it together, this interpreter, or on your order? Well, I'll have to see what's missing there...
This is not a commercial project. My role was reduced to active testing of the solution.
If you can figure it out, it will be good for the community.
And in general, so far I have not seen, what would really categorical signs would give an increase - but, tried a long time ago, I have predictors that describe local situations, as if categorical, but did not come.
no multiclass, as far as I understand
No regression yet.
Well if I do, it will be a parser of trained models from python to mql. I'm not burning yet, but may need it.
It doesn't matter, I can save models for python as well.
What's the point of regression?
It may be useful for models that work for setting stops. Sometimes it may be necessary to predict the MA through a dozen bars :)
Which of your features/transformations give good results?
The result of predictor's value depends on the target :) I'm currently conducting an experiment on the selection of the best quantization levels. Those predictors that didn't pass the minimum threshold are filtered out. It is too early to tell yet, but the first results are positive. The process is long in a single thread - more than a day. I need to use more criteria to evaluate quantum levels - I will do that - the idea is to dig where there is a signal. Further I will take more plots, filter sample and learn only where there is response - probably already genetic tree will work - to get leaves.
I made interesting thing, I can transform any dataset, marked or unmarked, improving it
This is interesting - you can try it on the one I posted the link. There's a mistake in some predictors there (when saved, they were written as int, not double - I removed my quantization and forgot), but for relative comparison doesn't matter.
By the way, if you need to calculate what is relatively heavy - I can calculate - now there is an opportunity.
This is interesting - you can try it on the one I posted the link. There is a mistake in a number of predictors (when saving were written as int, not double - I removed my quantization and forgot), but for the relative comparison does not matter.
By the way, if you need to calculate what is relatively heavy - I can calculate - now there is an opportunity.
I was looking for a very concise development of my approach and I stumbled upon something interesting... Or rather, it's not that I didn't know before, I just didn't think to use it... and then somehow the puzzles came together
It is not a panacea, but it gives interesting results. Later will be seen.
I was looking for a very concise development of my approach, and I stumbled upon something interesting... Or rather, it's not that I didn't know before, I just didn't think to use it... and then somehow the puzzles came together
It is not a panacea, but it gives interesting results. It will be seen later.
I am waiting with interest!
Spread cannot be beaten after simple decorrelation, but model is more stable on new data without spread. Any model that is overfitted for series, pours without a spread on n.d., but is much better on the tray than the first one (it works with a spread as well). This clearly shows the retraining to serial and nothing else. I know it's hard to understand, but it is 🤣 If you look at the pictures again, you'll see higher distribution peaks, and maybe tails, on the first one. That's seriality, volatility, whatever. It changes almost immediately on the new data, hence the overfit. The second bottom picture doesn't have that, it's all that's left, and in that garbage you have to look for an Alpha that beats the spread. Just look at your data and at least remove the serialization, or somehow transform it to remove the tails. And then look at the class distributions of what's left, whether there are normal cluster groups or complete randomness like mine. That way you can even visually see if the dataset is working or garbage. And then you can mix validation with trayn, it won't affect anything. And you say "just a picture".
You have to, Fedya, you have to!
Working day and night without sleep or rest ))))
Are you robots?
Working day and night without sleep or rest ))))
We have to, Fedya, we have to!
:))))