Machine learning in trading: theory, models, practice and algo-trading - page 1857
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
show pictures of kamins and kochu clusters, so unclear
kmins
kohonen
the prototypes look pretty similar but what can you expect with only three clusters
kmins
kohonen
If you look at the prototypes, it's pretty similar, but what can you expect with only three clusters
Damn, I should have scaled up the graphs, it's hard to see
I should try to write shorter TC, I'll get to it later. So yes, it's similar... but at the level of trading logic there may be some differences
Check that the rising and falling cohonen clusters do not go beyond +-0.001 on the first hour in one of the directions. And in general (on the average) the vector of price movement is already clear by the 1st o'clock (will it be directed or reversed).
You should also look for different pairs of neighboring hours. It's clear, that somewhere it will be better, somewhere worse.
I have to try to write the TS shorter, I'll get to it later. So yes, it is similar... but at the level of trading logic there may be some differences
The rising and falling cohonen clusters do not go beyond +-0.001 on the first hour in one of directions. And in general (on the average) the vector of price movement is already clear by the 1st o'clock (will it be directed or reversed).
You should also look for different pairs of neighboring hours. It's clear that somewhere it will be better, somewhere worse.
You mean filtering by hours that you may trade in and hours that you may not?
Do you mean filtering by hours in which it is possible to trade and in which it is not?
There are different clusters for different pairs of hours. Some signals are different for 1-2 hours, others for 5-6 hours. Some of them are not predictable at all
It's based on seasonal cycles and volatility clustering. Transitions from session to session are interesting, etc.
You can take more than 2 hours
and the perseptron gives worse results, but smoother
>>> clf.score(X_train, y_train)
0.7438271604938271
>>> clf.score(X_test, y_test)
0.7407407407407407
for different pairs of clocks different clusters. For 1-2 o'clock some signals, for 5-6 o'clock others. Some are not predictable at all
because it is all based on seasonal cycles and volatility clustering. Transitions from session to session are interesting, etc.
It is possible to take more than 2 hours.
And if we take segments of unequal size, but, say, decomposed into ZZ?
Have you tried to select more clusters and binary predict a specific cluster?
What if the segments are not of the same size, but, say, decomposed into ZZ?
Have you tried to select more clusters and binary predict a particular cluster?
The number of clusters does not change the quality of classification
Nope, I'm not into zz
the number of clusters does not change the quality of classification
No, ZZ is not for me.
Bravo Maxim, I give you a standing ovation for ZZ. ZZ is a dead end on all fronts.... I am serious. Proven more than once and all because it has no value on the zero bar, although attempts to create one have been a success, but only in the NS.
Otherwise Maxim very interesting study. I think you can do binary options at once with such a system. I think they will be wondering why you are so sagacious. No one will even guess that you're armed with a powerful mathematical apparatus, which breaks their BCs. ....
the number of clusters does not change the quality of classification
No, zz is not for me.
Very interesting - I need to try it, can you post the code, that in python?
Very interesting - I have to try it, can you post the code that is in python?
The trained models can be exported to the metaque at once, as ready-made functions. One tree or forest
and there you can do whatever you want in the tester, apply different strategies