Machine learning in trading: theory, models, practice and algo-trading - page 369
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
By the way, I checked the correlation with periods from 10 to 60 (6 pieces) on the EURUSD M1 chart, to the exit (I do not have a zigzag, but something close).
-0.00,0.01,0.00,0.01,0.01,-0.01
The correlation ranges from -0.01 to 0.01, i.e. there is no correlation at all.
However, your Expert Advisor shows profit. So you have to manually place trades looking at trendlinearre and making up certain rules based on chart movements. I believe it is much easier to write an ordinary Expert Advisor that will work by these rules.
Now, at the output, feed the increments of the same regression according to certain rules, with an offset, or several regressions (several outputs)... and the correlation will be normal. And feed more regressions with different periods to the inputs. MLP will build a cool regression model inside itself, like garch, and everything will be fine :) But in general, you need a more advanced neural network, LSTM, for example.
mine shows profits because it's run through optimizer, it's dirty results :) you can say that the fit, which will not work very long on forward (well periods will)
After all, on the chart, in addition to points obtained from the zigzag (or other method), there are dozens of moments / bars, when the trade will be profitable. And NS tries to adjust the trade only for this one variant of training.
And the example with trendlinearreg shows it well.
It seems to me that finding the right target markup is even more of a problem than finding good input data.
After all, on the chart, in addition to points obtained from the zigzag (or other method), there are dozens of moments/bars, when trades will be profitable. And NS tries to adjust the trade only for this one variant of training.
It is clearly seen in the example of trendlinearreg.
That's why NS has to be used as a part of a system, as a filter, or as an ensemble of different NSs.
All MO is based on the fact that the input variables must correlate with the output variable.
Otherwise, there is no point in ALL MO models.
You are seriously mistaken. Correlation is only LINEAR dependence, y = kx, even a trivial XOR dataset will give zero correlation of individual features with a target, nevertheless for a non-linear classifier easily solvable.
I've read THIS three times - it's hard to understand these scraps....
So?
I can take a multiple regression in which one or more (part of) the input variables will have a correlation with the output close to 0 and yet the model will give a high prediction accuracy.
So?
And if you remove these variables, the dimensionality of the problem decreases and the accuracy increases.
So what?
What is the point of your post?
The issue of discarding "unnecessary" variables solves the problem of reducing the dimensionality of the model.
For DM also increases the accuracy of model prediction.
For NS regarding accuracy - I do not know.
Read THIS three times - understand these scraps with difficulty....
So what?
I can take a multiple regression in which one or more (part of) the incoming variables will have a correlation with the outgoing one close to 0 and yet the model will give a high prediction accuracy.
So?
And if you remove these variables, the dimensionality of the problem decreases and the accuracy increases.
So what?
What's the point of your post?
chee, chee... oops! no way!...
Stop checking mister, we're not in a basement)))
You said that the chips should correlate with the target, those that don't correlate can be thrown out, I'm telling you that's not true, take XOR and check, there will be no correlation and the chips are important because the relationship is NOT linear, that's all, correlation only catches the linear component of the relationship.
what, what... oops! no way!...
Stop checking mister, we're not in a basement))))
You said that chips should correlate with target, those that don't correlate can be thrown out, I'm telling you that's not true, take XOR and check, no correlation and chips are important because the relationship is NOT linear, that's all, correlation only catches the linear component of the relationship.
Give me an example where the linear correlation is 0 and the non-linear dependence is strong.
Give me an example in which the linear correlation would be 0 and the nonlinear dependence would be strong.
I said XOR dataset
I said XOR dataset.
Do you have an example?
Show rows of incoming and rows of outgoing data - publish