Machine learning in trading: theory, models, practice and algo-trading - page 1437
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 couldn't figure it out for a long time, but it looks like it. Alglib forest is prone to overtraining when the sample size increases, because it is a nonstop bida. So, with moderate sample size can give good generalization ability, but with increasing sample size, the number of splits is off the charts and there without pruning you get just memorization. That is, when you increase the sample, pruning is necessary.
I did not see how it works in the new version on their site. Perhaps this flaw is exactly what is fixed
I couldn't figure it out for a long time, but it looks like it. Alglib forest is prone to overtraining when the sample size increases, because it is a nonstop bida. So, with moderate sample size can give good generalization ability, but with increasing sample size, the number of splits is off the charts and there without pruning you get just memorization. That is, as the sample increases, pruning is necessary.
I did not see how it works in the new version on their site. Perhaps this flaw is fixed.
Pruning should control the completeness, i.e. you have to cut it to the sample coverage of not less than 0.5-1%.
Why are you torturing the software from the last century, on the cyber-forum offered a variant five times faster. The author of NeyroPro admitted that he gave up positions, for a couple of decades, now they are writing code more optimal.
The multiplayer in C, not bad, not bad, our Maxim needs to feed it, he was looking for where to plunder MLP, and here is pure C 50 lines, although it is not his level yet.
No returnees from now on, I've sent you the best way, read it at your leisure)
All right, returns or any indicators are not necessary, the MO is able to find the dependence in the net price. I will be strict but fair, like in real life, or at work in the office, in relations with the Boss-subordinate.
I couldn't figure it out for a long time, but it looks like it. Alglib forest is prone to overtraining when the sample size increases, because it is a nonstop bida. So, with moderate sample size can give good generalization ability, but with increasing sample size, the number of splits is off the charts and there without pruning you get just memorization. That is, when you increase the sample, pruning is necessary.
I did not see how it works in the new version on their site. Probably, this defect was fixed.
Limit the depth yourself - put a counter (depth or number of examples in the sheet) and when exceeded, finish the division. In my experiments it did not lead to improvement on the feedback, still the same 50+-5%.
At least in that process I saw both stationarity and the presence of mutual information with the original row. There are some outliers, which can also somehow be fixed, but you see what you need to do.
The formula is simple, I rewrote it on mql.
I have also been pondering for a while about using mutual information, I think it makes sense, maybe even a grail.
All right, returnees or any indicators are not necessary, the MO is able to find the dependence in the pure price. Let me also in private your researches, I promise I will be strict, but fair, all as in real life, or at work in office, in relations the Boss-subordinate.
No, Kesha, in real life and here on the forum you don't have enough authority to share anything with you. Work on that.
No. There just rewritten functions I guess for speed. Depth there is still to the winning/last split.
Limit the depth yourself - put a counter (depth or number of examples in the sheet) and when exceeded, finish the division. In my experiments it did not lead to improvement on the feedback, still the same 50+-5%.
I have no idea how it works, but it says that produces an order of magnitude less wood, ie in fact should retrain less because the number of options is less, albeit in the same way at full depth
No. It's just rewritten functions I guess for speed. The depth is still up to the winning/last split.
Limit the depth yourself - put a counter (depth or number of examples in the sheet) and when exceeded, finish the division. In my experiments it did not lead to improvement on the feedback, still the same 50+-5%.
So you probably use returns, as everyone is desenformed forex demotivators alyosha and wacky wizard, and returns are independent, they already have no information, no levels or trend lines, pure SB.
I, too, have been thinking for some time about the use of mutual information, I think there is a sense in this, perhaps even a grail.
It's a very correct way of thinking, at least... the lib that I successfully rewrote from SI is just about that.