Machine learning in trading: theory, models, practice and algo-trading - page 939
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
That's what I'm saying.
To find the coefficients you have to see the result first.
And I'm telling you, it fits the prediction.
Right?
Of course it does. How else can you teach? Teaching requires an input and an output. With a teacher - without a teacher, in any case, the output is needed.
They won't give us what we need.
Of course. How else can we teach? Teaching requires an entrance and an exit. With or without a teacher, you need an entrance.
Here is a quote:
"Another variant of adaptation is possible, in which the reference signal is not used. This mode of operation is called blind adaptation or unsupervised learning."
Here's a quote:
"Another variant of adaptation is also possible, in which the reference signal is not used. This mode of operation is called blind adaptation or unsupervised learning."
So what? Did you look at the structure of learning without a teacher? In short - absolutely no different from the teacher. An ordinary system with an OS is a teacher.
So what? Did you look at the structure of learning without a teacher? In short - absolutely no different from the teacher. An ordinary system with an OS - that's the teacher.
Has anyone been able to achieve an error of 0.2 or 0.3 on the OOS? Moreover, it often works on OOS
But the difference of 2-2.5 times as much as trayne strains me
I can not understand when to finish development and start practice ))
I can't find news with day+hour combination, probably because of time change factor - to winter/summer...
Maybe some predictors are interfering...I made a grouping by hours.
Now time grouping by predictors is much better!
Screenshot is a test sample on all predictors without selection, if selecting, the result may be better.
Group 2 and 4 don't really work out, maybe they can be switched between them, but groups 1 and 3 are pretty good.
Made a grouping by hours
Now time grouping by predictors is defined much better!
Screenshot is a test sample on all predictors without selection, if you select, the result may be better.
Group 2 and 4 are not very good, maybe they can be re-sampled, but group 1 and 3 are not bad.
Try adding volatility readings to each clock? Or remove the hours and leave the volatility, which is cyclical by session
or grouping by trading sessions.
globally it should be 0.5, but quarterly +- there should be positive outliers
Note the quarterly cycles in forex
try adding volatility readings to each clock? Or remove the clock and leave the volatility, which is cyclic by session
Of course, you can use the indicator, but I immediately went further and searched for the cause of volatility - statistical news - I assumed that I would see it when breaking it down to days + hours, but it did not work - perhaps I need to consider the translation of hours for correct grouping, because I have Russian time, and news from America have a stronger effect on the ruble than our statistical...
Globally, it should be 0.5, but on a quarterly basis +- should be positive excesses
Note the quarterly cycles in forex
Quarterly is interesting, maybe that makes sense, we'll see, thanks.
However, the longer time intervals the less data there will be to measure - may be a pure fit.
Of course, you can use the indicator, but I immediately went further and searched for the cause of volatility - statistical news - I assumed that I would see them when breaking them into days + hours, but it does not work - maybe I should take into account the time translation for proper grouping, because I have Russian time, and news from America has a stronger effect on the ruble than our statistical.
Quarterly - interesting, maybe it makes sense, let's see, thanks.
However, the longer the time intervals the less data there will be to measure - could be a net fit.
Was referring to the fact that the patterns change every quarter, usually. The 7-years are still clearly traceable, but that's too positional
Maybe there are other periodicities that can be determined somehow fractally, I haven't done that, I need to look for information.
What I meant was that the patterns change every quarter, as a rule. The 7-years are still clearly visible, but this is too positional.
It's possible that somehow fractal patterns can be defined there, I haven't studied them, I have to look for information.
I.e. they change regardless of history, i.e. Q1 2016 is not like Q1 2017?
And fractals, so I have almost a fractal system for measuring price fluctuations in the range of 1 hour, 4 hours, 1 day, 1 week, 1 month. The planned fluctuation scale is calculated and we look where the price is at the moment (at what level).