Machine learning in trading: theory, models, practice and algo-trading - page 1037
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
There is a regular one.
Hello!
Does anyone know a smart extractor of quotes from mt4 to txt or csv file
inreal time mode.
........
But thanks anyway.
P.S. I got some help from good people.
Not something, but a times more complicated and inefficient analogue
and he himself can not fully explain what he screwed up and why :)
It's hard to say what's there, your article RANDOM DECISION FOREST IN THE TRAINING
I studied it last night, of course there is not much information, but I was very impressed by the example of work... Probably did not need to be posted! I spent half the night admiring the pictures of the tester, just looking at it! )))
If i'm serious, the machine learning itself seems to work, but the problem is in the input data - the machine needs to learn different parts of the price data, separately for flat or sideways movements, separately for trend movements, and the idea of choosing indicator parameters i still don't like - the market is constantly changing and the chosen indicator parameters are the same game - guessed or unguessed
as a first step, i think it is necessary to teach the car, for instance - if it was a trend day, then it will be a sideways trend - let the car learn to identify that moment at least - it's real machine learning
so it's like this
it's hard to tell what's out there, your article RANDOM DECISION FOREST IN THE TRAINING WITH CONNECTION
I studied it last night, of course there is not much information, but I was very impressed by the example of work... Probably did not need to be posted! I spent half the night admiring the pictures of the tester, just looking at it! )))
If i'm serious, the machine learning itself seems to work, but the problem is in the input data - the machine needs to learn different parts of the price data, separately for flat or sideways movements, separately for trend movements, and the idea of choosing indicator parameters i still don't like - the market is constantly changing and the chosen indicator parameters are the same game - guessed or unguessed
as a first step, i think it is necessary to teach the car, for instance - if it was a trend day, then it will be a sideways trend - let the car learn to identify that moment at least - it's real machine learning
like this
there is a link to a whole book for more details :)
Not something, but a times more complicated and inefficient analogue
And he himself can't explain up to the end what he screwed up and why :)
I just do not see the point in explaining something to someone who managed to confuse the threshold value from some modification with R parameter from AlgLib, which actually just divides the sample into teachable and test one.
Profit and "inefficient analogue" are still combined.
I modified the forest from AlgLib, so that it counts involved predictors. I don't want to disclose the list of predictors, because "they don't deserve it yet", but the number is saved.
I just do not see the point in explaining something to someone who managed to confuse the threshold value from some modification with the parameter R from AlgLib, which actually just divides the sample into training and test ones.
Profit and "inefficient analogue" do go together, though.
I modified the forest from AlgLib, so that he was counting the predictors involved. The list of predictors I do not want to disclose, because "do not deserve more", but the number is saved.
49
No one here understands you, including me. Since you know how to write code, but you can't express your thoughts in letters
What threshold value and r, I haven't written anything at all
Why post the library without a description, and then write that "you don't deserve it"?
I just do not see the point in explaining something to someone who managed to confuse the threshold value from some modification with the parameter R from AlgLib, which actually just divides the sample into training and test ones.
Profit and "inefficient analogue" do go together, though.
I modified the forest from AlgLib, so that it counts the predictors involved. The list of predictors I do not want to reveal, because "do not deserve it yet", but the number is saved.
Did you happen to modify the forest so that you could prune trees in it? It would be interesting to try.
Maxim Dmitrievsky:
what threshold value and r, I didn't write you any such thing at all
The forest doesn't give out class membership probabilities, so these inequalities are nonsense
>< 0.5 and that's it, there's no other way. And another question is what is better, binarized traits and outputs or not.
you can divide classes from 0 to 100, there is no differenceah, well yes
The result of all classification algorithms included in ALGLIB package is not a class to which an object belongs, but a vector of conditional probabilities.
But this is not much consolation. There will be fewer signals and the efficiency will not necessarily increase. I, for example, did not increase, everywhere I put 0.5 threshold now
What's more important is the comparability of errors on the train and the oob.
I'm kind of an algleb too)
And then I realized that the "threshold" is named parameter R from AlgLib.
Reading source code gives much more than reading theoretical articles. The programmer must read the source code on which the program execution depends.
What about the previous posts?
At first I thought modifications were being used, of which there are many. True, the concept of "weight" is used there, not "threshold". Well mixed up... But then this:
And then I realized that the "threshold" is named parameter R from AlgLib.
Reading source code gives much more than reading theoretical articles. The programmer must read the source code on which the execution of the program depends.
I gave a quote from the AlgLib website:
"the result of all the classification algorithms included in the ALGLIB package is not the class to which the object belongs, but a vector of conditional probabilities."
i.e. confirmed your words that the output is probabilities. These are of course pseudo-probabilities, but still. I did not study in detail how they are counted, but by logic from the words "probabilities" there is only one name.
By any chance did you modify the forest so that you could cut trees in it? It would be interesting to try.
I quoted from the ALGLIB website:
"the result of all the classification algorithms included in the ALGLIB package is not the class to which the object belongs, but a vector of conditional probabilities."