Machine learning in trading: theory, models, practice and algo-trading - page 1537
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
arguments?
What arguments are needed - parameter 6 by default, follows from the settings table.
Strange, because the depth of the tree depends more on the complete links between predictors, that's why I'm surprised that from these two independent links, the total complexity in the region of 200 you get a good model, judging by the graphs.
What arguments are needed - parameter 6 by default, follows from the settings table.
Strange, because the depth of the tree depends more on the complete links between the predictors, so it is surprising that from these two independent links, the total complexity in the region of 200 turns out a good model, judging by the graphs.
That's why I'm saying the data is out. Forest retrained on them (maybe be able to fix this feature of the forest, by categorizing the features). Now put 6 - re-training on trayn goes great, acurasi at 0.9.
6 is a little too much for mine, 2-4 is normal
The analogy with the forest is direct, there the depth of the trees is not limitedThat's why I say that the data is taken out. Forest re-trained on them (maybe be able to fix this feature of the forest, by categorizing the features). Now put 6 - re-training on trayn goes substantial, acurasi under 0.9.
6 for my too much, 2-4 normal goes
The analogy with forest turns out straightforward, there the depth of trees is not limitedI do not understand, you said earlier that you have predictors in the form of increments, then how do you want to turn them into categorical predictors?
A lot doesn't essentially change from tree representation - I dissected trees, there are many individual leaves of the tree just combined into one long leaf essentially, and very much of those leaves are duplicated or have meaningless intermediate values/links that can be pruned. Generally I've seen on my sample that tree depth affects the number of trees, and you can get the same result on a tree of 4 splits.
I don't understand, you said earlier that you have predictors in the form of increments, then how do you want to make categorical predictors out of them?
A lot doesn't essentially change from tree representation - I dissected trees, there are many individual tree leaves just combined into one long leaf essentially, and a very large portion of those leaves are duplicated or have meaningless intermediate values/links that can be pruned. Generally I've seen on my sample that the depth of the tree affects the number of trees, and you can get the same result on a tree of 4 splits.
First split into categories, for example 20 ranges-categories. Then vanchot encoding (via dummy attributes) or something else, I haven't decided yet. In the end each feature will be binary or something like that.
The more different values for the forest, the more overtraining. As the training sample increases, overtraining increases. For catbust, it doesn't. So for the forest, try to reduce the number of choices for continuous traits by categorizing them. I'm not sure if this will help, but we'll see.
I don't know if it'll help or not.
First break it into categories, for example 20 ranks. Then vanchot encoding (via dummy attributes) or something else, I haven't decided yet. In the end each feature will be binary or something like that.
I don't know, it's more about speeding up data processing, not fragmenting such predictors, not comparing each other in a chain of leaves, I didn't see any worthwhile effect, unfortunately. And the logic is that these are not comparable values, this is what vanchot encoding, combined into a group in order to equalize random selection.
the more different values for the forest, the more overtraining. As the training sample increases, overtraining increases. The catbust does not.
The dependence of the sample on training there is also not unambiguous - I did half a year or earlier similar studies. More likely there is a dependence on the data, which should be comparable.
A simple and interesting approach to describing patterns for MO
https://www.quanttrader.com/index.php/a-simple-algorithm-to-detect-complex-chart-patterns/KahlerPhilipp2019
mega-primitive, there are much more accurate ways to describe a pattern
mega-primitive, there are ways to describe the pattern much more accurately
such as?
I am willing to pay a reasonable amount for the Grail, confirmed on the real (test reports do not interest me) for at least 3 months of work.
I believe that the real value of the Grail = the sum of the current equity of the trader. I.e. in the account of $ 1000 equity, then the TS is worth as much. If anyone has a Grail, based on neural network technologies and/or physical and mathematical models, real stats, and is ready to sell it, please do not hesitate to message me in person and we will discuss it.
Who really earns on the market will not sell their technology for millions of green, maybe for hundreds of millions... I'm trying to compare a kid in a sandbox and a gold-mining company... The kid wants to dig gold by his scoop but he can't make it... Can we use a different shovel? Ask the other children in the garden who have such a shovel? Most likely someone "will", children are often dreamers:)
I don't know, it's more about speeding up data processing, not splitting such predictors, not comparing each other in the same chain of leaves, I didn't see any worthwhile effect, unfortunately. And the logic is that these are not comparable values, they are vanchot encoding grouped together to equalize random selection.
The dependence of the sample on learning there is also not unambiguous - I've done half a year or earlier similar studies. More likely there is a dependence on the data, which should be comparable.
In that article in English just about this, yes, that would not compare values of one variable with each other, when there are many - this only leads to overtraining
Maybe not from the length but some other, I tell you what I see. I increase sampling - it becomes prettier on Trayne, worse on the test. Although generalization should increase with increasing tray size, but it's exactly the opposite for forests.
Such as?
dtw, spectral analysis... a bunch...
I managed to create an algorithm that knows how to see the same patterns regardless of their magnitude, so the algorithm looks at one chart and sees a pattern on both the 1-minute and weekly charts, looking at only one chart, and it can really make predictions, but there is still a lot of work