Machine learning in trading: theory, models, practice and algo-trading - page 435
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 on what principle do simple NSs (simple MLP) make a prediction?
I think on usual correlation - because weight of connections between neurons grows with number of repetitions of signal along this line with coinciding response of NS, if the line was in + and in - it will remain about 0 - and it is, in fact, usual averaging. Then using these weights we find the similarity of the input combination of predictors to the average on the training period.
Approximates the f-ions
the only option is to ask the stableman for help) he will teach how a real man should trade.... not patterns and science are important, but courage and strength... and you need a real Chechen beard... then the market will not stand against an inflexible and principled warrior.....
hutch style trading rules..........
The stablemate is a miserable man who thinks he is a guru and has decided to overcome the market through his psychology. This is the first stage of ignorance, there is much more to come...
i wish him luck turning from a nub into something more capable
A stablemate is a miserable man who fancies himself a guru and decides to overcome the market through his psychology. This is the first stage of ignorance, there is a lot more to come...
I wish him good luck to turn from a nerd to something more capable.
But he is nevertheless your teacher, he wrote somewhere, so something useful he tells you, if you pay him money.
But he's still your teacher, he wrote somewhere, so he's telling you something useful if you pay him money.
He's a liar, he must have been dreaming about it, I've already posted the correspondence... he's a sick man. That is, he thinks he is teaching me something.
Of course, I did not pay him any money. This jerk stuck to him like a bath sheet, after that I just banned him.
He's a f...t, he must have dreamt it, I've already discounted the correspondence... he's a sick man. I.e. he thinks he is teaching me something.
At the prank looks like someone who just has nothing to do, it is unlikely that he writes seriously, think about it, what the fuck, a real "horse" to do here? But nevertheless, as the saying goes, "residue remains" and you are now his disciple, associated with the stable boy and his teachings on courage and adult trading.
It looks like a prank, someone who just does not give a shit, it is unlikely that he wrote seriously, think about it, what the fuck, a real groom to do here?
I don't care, he's just wasting my time with his bullshit ) ask him what he's even doing here, I'm not interested.
Once again: he is not my friend, not my brother or brother-in-law, not my teacher, I have no idea why this scum thinks he is my teacher and even insults me.
If I understand what you mean, I sift out all adjacent ones for 20 bars after the variant I have found.
No, it's more like finding a pattern, dividing it into parts, and then seeing how those parts worked out as predictions, and then running through all the parts and seeing the average error... almost like a neural network
Or, we look for several consecutive patterns and then see how all these patterns worked out; if the mean error is low then we can trust the next prediction.
So there will be a factor of interaction of one pattern with another, so we analyze something like context, i.e. in what positions these patterns are relative to each other, in theory this should improve the quality of the prediction. On the other hand how do I separate these patterns, they can be represented as one big one if the correlation with the current chart is good...
This is what I was doing, it may be useful, with slopes of regressions... in the visualizer and it shows the prediction. The red chart shows the real chart, the green one - with a changed slope angle, according to which the pattern was found, then the forecast is transferred to the red one.
You can set the Forecast accuracy less and the scaling stop too, otherwise the quotes may not be enough. It builds a lot of timeframes with a multiplier.
Sometimes it predicts great, sometimes it blows it out of the water.)
No, it's more like finding a pattern, dividing it into parts, and then seeing how those parts worked out as predictions, and then running through all the parts and seeing the average error... almost like a neural network
Or, we look for several consecutive patterns and then see how all these patterns worked out; if the mean error is low then we can trust the next prediction.
So there will be a factor of interaction of one pattern with another, so we analyze something like context, i.e. in what positions these patterns are relative to each other, in theory this should improve the quality of the prediction. On the other hand how do I separate these patterns, they can be represented as one big one if the correlation with the current chart is good...
This is what I was doing, it may be useful, with regression slopes... you run it in a visualizer and it shows the prediction. The red chart shows the real chart, the green one - with a changed slope angle, according to which the pattern was found, then the forecast is transferred to the red one.
You can set the Forecast accuracy less and the scaling stop too, otherwise the quotes may not be enough. It builds a lot of timeframes with a multiplier.
it is interesting to look but #include <MT4Orders.mqh> is missing and if it is commented out ScaleInvarianceMultiTF (EURUSD,M1) array is out of range in 'ScaleInvarianceMultiTF.mq5' (79,55)
https://www.mql5.com/ru/code/16006
Out of range - the history of quotes is not enough, set the scaling stop setting less, 50 for example.
That is, if you take a pattern of 100 bars on the minutes, then it will need 100*50 bars of history to build all the synthetic timeframes, and it will need 100*1440 there :)
https://www.mql5.com/ru/code/16006
Out of range - the history of quotes is not enough, set the scaling stop setting less, 50 for example.
That is, if you take a pattern of 100 bars on the minutes, then to build all the synthetic timeframes it will need 100*50 bars of history, and it will need 100*1440 there :)
Does it look for 1 similar variant or does it average over several ones? Apparently it finds the best one. I think I should look for an average forecast for 10 or even 100 variants (the exact number should be determined by the optimizer).