Machine learning in trading: theory, models, practice and algo-trading - page 1287
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
I'm actually self-taught. Mostly I deal with websites.
And when I'm not, I do the MO. As a hobby).
And with the tree once you have to sit down and figure it out. Then you can remake it any way you want.
The old one is simple, and the new version has something screwed up (probably, calculation gas pedals, but the code became more complicated) - I didn't even bother to look into it.
I'm figuring it out right now, looking for the place where I can change the number of samples.)
I'm sorting it out right now, looking for the place where I can change the number of samples)
//--- to split or not to split
if(idxbest<0)
Well, just above you can also block loop splitting node (to speed up), but you can leave it as it is, it will find splitting, but it can be ignored.
But also understand us - we were "yelling" only because we were waiting for you to come and show us all how to predict the price through returns and log-returns.
Burn!
What is it that angered you, my friend? What is this sargasm from? You don't know how to turn a price into a log-return? I'm not saying these returns are well predicted, alas no, but NOT-stationarity is a bit different. Good luck to you, don't worry.
What is it that angered you, my friend? Why the sargasm? Don't you know how to turn the price into log returns? I'm not saying that these returns are well predicted, alas no, but NOT-stationarity is a bit different. Good luck to you, don't worry.
I see.
And this one's gone...
Better here is to add a stopping condition on the number of examples
//--- to split or not to split
if(idxbest<0)
Well, just above you can also block the node split loop (to speed up), but you can leave it as it is, it will find the split, but it can be ignored.
somehow the error rises sharply even at <2
somehow the error rises sharply even at <2
If samples=10, then tree error is less, if 100, then more. It is possible to find an optimum. even 1000 if there are 1000000 rows.
is good. And without it on trayne tree to error=0 learns, i.e. is retrained. I have a tree on trayne about 30%, the forest lowers to 10 or less.
If samples=10, then tree error is less, if 100, then more. You can pick an optimum. even 1000 if there are 1000000 rows.
somehow I have not yet appreciated the benefits, the error can be increased through r
Starting with 100k strings I do not learn anything, the error is in the region of 0.5. Up to 10k works fine, but not for long :) The problem is more in non-stationarity, searching for predictors (if they exist). Any differences (increments, cumulative sums, etc.) do not lead to success. On some uniform chunks or periodic f's it works perfectly. As soon as I see serious market changes, it fails. As a result, I concluded that predictors should be different, but I have no idea what they are.)
For example if I wanted to use an appropriate calendars for my forex broker, I would have to buy them with an aptitude for calendars. I think it makes sense to try it later.I see.
And this one leaked...
What's clear, and who leaked? Are you in a normal state?
I also tried synthetic timeframes like Renko or zigzags on ticks, with different periods (without reference to time), hoping that there are some interesting distributions compared to the usual time chart. +- the same, I did not find anything particularly interesting.
i.e. non-stationarity is not killed by all such bullshit and regularities are not better found
By the way, in mql5 added api news calendar, maybe it's still raw, I haven't tested it. I think it makes sense to try it later.
Yes, I read it too. It will be possible to test it.