Machine learning in trading: theory, models, practice and algo-trading - page 2805
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
Sometimes it is useful to superimpose the balance chart (blue) on the price chart (orange)
In this case it is clearly visible, for example, that the model has been earning only on a falling market for 10 years
The more refined model similarly learns to earn mostly on dips, but draws down less on rises and consolidations. Because the market has been falling for all 10 years, the training sample is biased
Are you sure it's randomly selecting the priors? I wasn't catbusting, I was looking at the code of basic bousting examples. All the predictors are used there. That is, the best one is taken. The correlated one will be next to it, but slightly worse. But at some other split levels or in correction trees, another of the correlated predictors may be better.
There is a separate parameter that allows to take only part of predictors for estimation - it is taken randomly - it increases the number of model trees, but in theory it speeds up training.
I would be sure if I had parsed their code, so I'm only talking about what they declared - there's a small chance I misunderstood the developers.
So what they're talking about is adding randomisation to split evaluation, by doing this they have improved learning.
That there is preprocessing to eliminate correlated predictors - I haven't heard about that.
I agree that another predictor may be better or more useful, but my goal is to train as many different models as possible. I want them to learn something.
So throw in a couple of informative formulae for me to test out.
I don't get it, should I throw it to you or to the dataset? If in dataset, how is it related to grouping similar or correlation, because the target is not involved in this process?
I don't understand, should I throw it to you or to the dataset? If in dataset, how does this have anything to do with similarity grouping or correlation, because the target is not involved in this process?
formulas for calculating good fiches to reproduce in python and mql. I'll pick up the tags myself.
just to look, because I only use increments.
I can upload bots for them after training.Sometimes it is useful to overlay a balance chart (blue) on a price chart (orange)
In this case it is clearly visible, for example, that the model has been earning only on a falling market for 10 years
The more refined model similarly learns to earn mostly on dips, but draws down less on rises and consolidations. Because the market has been falling for all 10 years, the training sample is biased
Is the price built on the principle of deltas between closing and opening trades or is it some kind of time rationing?
Is the price built on the principle of deltas between closing and opening a trade or is it some kind of time rationing?
yes, at the moment of trades it is fixed
formulas for calculating good chips to reproduce in python and mql. I will pick up the labels myself.
just to look, because I only use increments.
I can upload bots on them after training.You know that the "goodness" of a feature is determined by its target.
I have described the principle of creation earlier. The code in MQL5 is all in MQL5, and there is no question of any conversion through a function - it is not possible to give a formula.
For example, the start time of the current segment ZZ(48) is often selected.
You know that the "goodness" of a feature is determined by the target.
I have described the principle of creation earlier. The code in MQL5 is all in MQL5, and there is no question of any conversion through a function - it is impossible to give a formula.
For example, the start time of the current segment ZZ(48) is often selected.
ah, well, mostly on standard indicators and their derivatives?
Yes, at the time of the transactions is fixed
Try to make a markup for buying and selling, selecting a model more balanced by the number of inputs, and then split the sample into two and make two separate models.
Try marking up to buy and sell, selecting a model more balanced by the number of inputs, and then split the sample into two and make two separate models each.
to train separately on a rising market and on a falling market? Then the signals will be averaged anyway.
Everything is logical, counter-trend trades are usually ineffective in trending markets, with some scalper exceptions.
I just showed you that you can get a better understanding of how the model works.