Machine learning in trading: theory, models, practice and algo-trading - page 896
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
and let me tell you that these models work exactly the same way as Reshetov's on OOS. Exactly the same.
i.e., in your case, NOTHING? i will tell you the same, more lenient 3rd grade parochial
i.e., in your case, NOTHING? say the same, more than the 3rd grade fullness of the church parish
Not this time, Maximka, not this time....... The magician was right. Data quality has increased by an order of magnitude, hence the quality of the model......
My two favorite magicians are friends again. What lies ahead for them)))
Not.... he hasn't apologized yet.... although there's nothing to apologize for yet....
No, I've never done it.
I don't really understand why it's needed, because, for example, a forest is already a universal classifier or approximator, and there's nothing to fix by hand
and single trees are rather weak and primitive algorithms.
The tree is more like a clear logic, which is formed after analysis, and everything else looks like a fitting for me so far.
I was interested in the tree as a convenient tool for taking into account different situations in the market, i.e. it is convenient to use one and the same pattern under different conditions and the selection of these conditions will either activate the pattern or not. It is a very cool tool for formation of complex patterns that can be easily programmed, but there is no convenient program.
And if the tree, gives a result in 90% of the available predictors, this is already more than a lot, then what is the point in NS or ensemble of trees? I noticed that if tree uses logical rules (>/</==) to select areas of predictor classification, then the result is better, I understand that the tree does not look through all available results, or because of the cut rules with little support are cut, in the end 5-10% of reliable recognition sample is lost.
And if the tree gives results in 90% of available predictors, then this is already more than a lot, then what is the point in NS or ensemble of trees? I noticed that if the tree uses logical rules (>/</==) to allocate areas of classification of predictors, the result is better, I understand that the tree does not go through all available results or because of the cut rules with little reinforcement are cut, resulting in a loss of 5-10% reliable recognition of the sample.
the point is that 1 tree just learns the sequence, with no chance of adapting to changing patterns (and in the market they always will)
ensembles have many random components that increase robustness on new data
Yes, cutting the tree is called pruning
You're a stoner, Maxime the Trickster doesn't give a damn about you at all.
He doesn't understand the depth of your soul.
You got into a fight with an automaton yesterday. Be careful with it)))) hilarious...
Why am I a magician? My posts don't magically disappear
just a statistician provocateur.
and i don't smoke
the point is that 1 tree simply learns the sequence, with no chance of adapting to changes in patterns (and there will always be some in the market)
ensembles have many random components that increase robustness on new data
yes, tree slicing is called pruning
How can we know how patterns will change? Can the network divide the sample into 100 parts and examine in those sigments not only the relationships of the predictors, but also the nature of the change in those relationships, i.e., the patterns? If it does that, then yes, it can infer the rules for changing a pattern, but everything I've read about it seems like perverse classification.
What I like about the tree - it's setting the hierarchy, in the program I'm working in now you can create a basic hierarchy, and already from it to do the automatic calculation (so far it's troublesome, but maybe I do not use all the tools, because many I just do not understand their name). For example to create ATC I use such questions, to which the TS should give answers to make a trading decision:
- Where am I? (Description of the price point).
- How did I get here? (Analysis of the movement from the conditionally opposite point to the current one)
- What can happen? (Calculation of probable events when they occur in the future)
- What to do? (Analyzing historical patterns given the answer to the first 3 questions)
- Is it worth the risk? (Analysis of possible losses from entering the transaction and possible profits)
That's what I want the network/tree/forest to answer these questions in the same sequence when making a trading decision.
How can we know how patterns will change? Can the network divide the sample into 100 parts and examine in those sigments not only the relationships of the predictors but also the nature of the change in those relationships, i.e., the patterns? If it does that, then yes, it can infer the rules for changing a pattern, but everything I've read about it seems like a perverse classification.
The forest does that, neural networks with crossvalidation and\or ensembles of underlying models too
+ various regularizations
Forest does that, neural networks with crossvalidation and\or ensembles of basic models too
About NS - maybe I don't understand something yet...
How does a forest do that? The forest just takes predictors randomly and looks for connections between them, then votes. It will either vote wrong/wrong on a sample under a stationary model depending on the resulting sample/predictor randomness, or its correct vote will be biased by that sample, assuming the patterns in the sample change. But where is the analysis of changing patterns? As I understand it, the forest is good when you don't know how good the predictors are, then in the absence of mutual reinforcement of predictors such predictors will either have no weight in voting or their weight will not be significant. Or do I have it all wrong?