Machine learning in trading: theory, models, practice and algo-trading - page 1927
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
random...
The fixture can be represented in the form of a log rule...
rule size - random
content of the rule - random
generate 1000 rules - send them to IMO as 1000 features
selected 1-5 good features if there are any, if not - we drop all of them.
selected features get thrown into "good features database".
and again we generate 1000 features, and so on
when there are over 1000 good features in the "base of good features", you will be able to use them to train a new model and see what you get.
Randomly generating rules without borders is like running from edge to edge in a bare field. Reasonable boundaries are needed, or at least look for 1000 rules in one logical and dimensional area, another thousand in another. a complete randomness is from minus infinity to plus, though) But the idea is ok)
To randomly generate rules without boundaries is like running from edge to edge in a clean field without boundaries. A reasonable boundary is needed, or at least look for 1000 rules in one logical and dimensional area, another thousand in another. full randomness is from minus infinity to plus all the same) But the idea is ok)
this is all a nuance, you can make a selection from good chips like breeding in genetics, and then a random search will slowly evolve into a directed .....
It is like in the song, "I shout out, but there is silence in response.
This is all a nuance, you can make a selection from good chips like in genetics, and then a random search will slowly grow into a directed .....
Karoroche you can rub it as much as you want and with what you want, who will do? And here it is like in the song - "I shout, but in response silence.
First at least define the zones of rules. I don't know which ones. What kind, how deep and wide are the searches. On one or all of the TF, and then a few series and the dimension flies away. I argue that we do not need seasonal search. It's hard to get started without it)
First, at least define the zones of the rules. Lag, indicator need not and what, the depth and width of the search. On one or all TFs, and it is a few then rows and dimension flies away simply. I argue that we do not need seasonal search. It's hard to get started without it)
You don't know what you need and what you don't. If you knew, you wouldn't have to search, that's the main idea, the machine will find solutions that you would never think of, and if you limit it, you'll get another stochastic
I do not know the clustering algorithm that creates the rules.
During clustering, many rows were distributed in different areas, a map was formed, which, as I assume, can be called through:
And then weighting each row to assign it to one or another center of the cluster. I just don't understand how the weighting of a single line occurs...
I already told you, interpret it according to the direct purpose of the tool, and you're trying to hammer nails with a flower.
Yes I am, already got the trees in training, for each cluster is different - there are leaves with good completeness and acceptable accuracy - in the evening I'll put together an EA and see what happens on the new data, what will be more effective - to use a single tree or a breakdown into 4.
I have decided to look at significant reversals of the market. Significant U-turns as a target. Thought it would be chaos, but no.
What's the rule for classifying a reversal as a significant one?
Well, that's interesting. Thanks for the tip.
Can you share the code for dummies, maybe I'll join the eRs?
Thank you, I managed to unload the clustering.
random...
The fixture can be represented in the form of a log rule...
rule size - random
content of the rule - random
generate 1000 rules - send them to IMO as 1000 features
selected 1-5 good features if there are any, if not - we drop all of them.
selected features get thrown into "good features database".
and again we generate 1000 features, and so on
when there are more than 1000 good features in the "base of good features", we will be able to use them to train a new model and see what happens.
So how is your method better than mine - collecting leaves is in fact new predictors created from existing data. You just need to build trees not only using comparison, but also transformation and combining levels of the target, in general you can implement it on the basis of a regular tree and drag leaves from there.
You don't know what you need and what you don't, if you did you wouldn't have to search, that's the point - the machine will find solutions you could never think of, and if you put it into your box you'll get another stochastic
You can search only in a certain space, and you can't do it without defining the search area.) It's a dumb machine, but it works.)
If there are of course ideas of searching for rules without rules for some regularities in multidimensional series or at least in one-dimensional ones, we should at least formulate how to approach the idea of generating rules without rules. Rules as well as regularities are infinitely many by the condition of the problem.
Anyway, I screwed on a catbusta instead of trees, and... the problem was not in the trees, but as always, in the head :z
method doesn't work as expected
Did you use Aliaksandr Hryshyn's class ?
So how is your method better than mine - collecting leaves is essentially new predictors derived from existing data. You just need to build trees not only using comparison, but also transformation and combining levels of the target, in general you can implement it on the basis of a regular tree and drag leaves from there.
Question in the chip / log rule generation
The question is about generating features / logging rules.
That's how the chips will be generated - we have to prepare a constructor in the form of basic rules.
For example, describe once how a price behaves in a channel and then just change channels and so on.