Machine learning in trading: theory, models, practice and algo-trading - page 3370
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
A pattern, for example, in wooden models, of which there are many, models, is a tree. Each wooden model, for example, CatBoost finds over a hundred trees, read patterns. For RandomForest I have statistics: up to 50 trees the classification error drops, and over 150 trees the classification error is stable, i.e. in the time series I processed the number of patterns does not exceed 150.
It is only a little bit left to formalise these 150 patterns to one common pattern. If there is A and there is B, then C will come at a certain time in such and such a duck, if not, the time and point will be different. There are only three different variants of C's development. Everything is decided by TIME! And this can be easily worked out first by hand, and then you can put it into a robot, if you want.
P.S. - not only the size of A already gives certain levels for B' s pullback (very often the backlash does not exceed 1-3 points). Similarly, B signals in advance about the price levels of C. There is a lot of maths everywhere.
As an example, a manual Expert Advisor automatically calculates levels from the movement we are interested in. I.e. we pull on the desired wave, and the future is already marked.
A pattern, for example, in wooden models, of which there are many, models, is a tree. Each wooden model, for example, CatBoost finds over a hundred trees, read patterns. For RandomForest I have statistics: up to 50 trees the classification error drops, and over 150 trees the classification error is stable, i.e. in the time series I processed the number of patterns does not exceed 150.
Rather, the path to every leaf whose response exceeds your threshold for opening a trade is a pattern.
For example:
1st split: 40 bar delta with current = 0,
2nd split: 30 bar=100,
3rd split: 20 bar =50,
4th split: 10 bar 100,
5th split: 1 bar = 0.
That's 2 tops with a trough. A pattern? Yes. The path to 1 leaf describes it.
If you have 20 leaves in your tree that exceed the threshold you are trading, you have 20 patterns in your tree that you use. Less successful leaves - also describe patterns, but they are random/loss.
Other 100 trees can use other chips/bars with different split values.
Forest will average (add and divide by the number of trees) the answers of all trees, we will get an average opinion about the situation.
Boost will simply add with a weighting/clarifying coefficient.
There is just alittle bit left to formalise these 150 patterns into one common pattern. If there is A and there is B, then C will come at a certain time in such-and-such a duck, if not, the time and point will be different. There are only three different variants of C's development. Everything is decided by TIME! And this can be easily worked out first by hand, and then you can put it into a robot, if you want.
P.S. - not only the size of A already gives certain levels for B' s pullback (very often the backlash does not exceed 1-3 points). In the same way, B signals in advance about price levels of C. There is a lot of maths everywhere.
As an example, a manual Expert Advisor automatically calculates levels from the movement we are interested in. I.e. we pull on the desired wave, and the future is already marked.
There is no need to formalise anything - everything is done for you.
Rather, the path to every sheet whose answer exceeds your threshold for opening a trade is a pattern.
For example:
1st split: 40 bar delta with current = 0,
2nd split: 30 bar=100,
3rd split: 20 bar =50,
4th split: 10 bar 100,
5th split: 1 bar = 0.
That's 2 tops with a trough. A pattern? Yes. The path to 1 leaf describes it.
If you have 20 leaves in your tree that exceed the threshold you are trading, you have 20 patterns in your tree that you use. Less successful leaves - also describe patterns, but they are random/loss.
Other 100 trees can use other chips/bars with different split values.
Forest will average (add and divide by the number of trees) the answers of all trees, we will get an average opinion about the situation.
Boost will simply add with a weighting/clarifying coefficient.
A tree is a tree and you can even print them out, as I recall.
Leaves, on the other hand, are a dark matter. It is at the leaf level that the algorithm, together with the designer's pens, seeks a balance between classification error and bias. It is possible to set the trees so deep that they describe the input set exactly - overtraining or overfitting. And the problem is to get classification with minimal error in a future where we don't know the input set. Leaves they are finite of course, but they need to be approximate, not exactly "finite". Therefore, a pattern is a tree.
A tree is a tree and they can even be printed out, as I recall.
But the matter with leaves is dark. It is at the leaf level that the algorithm, together with the designer's pens, searches for a balance between classification error and bias. It is possible to set the trees so deep that they describe the input set exactly - overtraining or overfitting. And the problem is to get classification with minimal error in a future where we don't know the input set. Leaves they are finite of course, but they need to be approximate, not exactly "finite". Therefore, a pattern is a tree.
I disagree. I gave an example of a leaf with a depth of 5 (the leaf is obtained through 5 splits and it is a pattern). If you get a leaf through 200 splits, it will give you an over-trained pattern that will perform poorly on new data. But it won't stop being a pattern. A tree is a set of patterns.
I disagree. I gave an example of a 5 depth sheet (the sheet is obtained through 5 splits and it is a pattern). If you get a sheet through 200 splits, it will give you a retrained pattern that will perform poorly on the new data. But it won't stop being a pattern. A tree is a set of patterns.
Let's start at the stove.
I'm considering two variants of trees: bagging and boolean (rf, ada) types.
Either of these rather large set of algorithms considers the input data POSSIBLY, without reference to neighbouring rows. that is why the training set is created by randomly selecting rows.
Within a single row, a selection of predictor values is made that will predict the class specified by the target variable (the teacher) with the highest probability.
The combination of some number of predictors from the set of input predictors and their values in a single row is fairly called a pattern.
Within a single line, the predictor values are selected that will predict the class specified by the target variable (teacher) with the highest probability.
The combination of some number of predictors from the set of input predictors and their values in one line can be called a pattern.
You misunderstand the stovepipe. It seems that you have never looked at the tree construction code.... There are no operations within one line there at all!!!, only with sets (with full or with batches).
In brief:A random/full set of rows passed to training, is sorted one by one for each predictor/column. Different splits on it are checked (middle/percentile/random), statistics for each are counted, and the best split is selected for the whole set of rows, not for one/each row as you suggested.
According to the best split, the set of strings is divided into 2 sets, then each set is sorted again and the best split is selected for each of the parts, etc. until the stopping rule is reached (by depth, number of examples per line, etc.)
You can see more details in the editor, you have the file:
\MQL5\Include\Math\Alglib\dataanalysis.mqh
ClassifierSplit() function and the one from which it is called.
You will understand it in a couple of hours and you won't have to talk about searching predictors by one line.
There is no need to formalise anything - everything has been done for you.
And what is the accuracy of your robot on the interval plus or minus a couple of three hours into the future? I'm talking about the pivot point that your AI will calculate into the future. With an error of plus minus 5-10 pips intraday on the same pound-dollar, and an error of +-1-2 bars on m5.
And your AI will be able to realise in time that the model (pattern) has moved not by price, but by time? This means that not the first, but the second or third calculation point will be triggered. :)
Coders are like children, if someone doubts the capabilities of a computer (like a machine will find everything), they immediately pout their lips and take offence. ;) And nobody can catch up that you will never see why you can't find a 100% working algorithm. And all because there is one small imperceptible detail, it is it that gives a signal that the proportions of the pattern in the future will shift by a certain amount. But you are Heniuses and Gods of code, it is not the honourable thing to pick something (to search everything by hand, counting on a paper or calculator). It is easier to code some bullshit at the input, get bullshit at the output, and then try to optimise it all.
A clear logical mathematical system was described more than 100 years ago (it still works like clockwork). Both in terms of algebra and geometry. But the inventors of their own lysapedes are still in abundance....
A clear logical mathematical system was described over 100 years ago (still works like clockwork). Both in terms of algebra and geometry. But there are still plenty of inventors of their own lysapedes.....
So show me your trade.
Questions like that will make everyone dizzy.
It's like school.
You have to start from far away and ask simple questions.
What pair do you trade on?
And show the last trade, entry exit with all parameters.
And then, how many trades a day you make and in this place there is a pause.
P.Z.
You should not ask more than two questions.
You can offend masters. And what if you start to do so?
P.Z.
I support Spartak.