Machine learning in trading: theory, models, practice and algo-trading - page 1543
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
It looks interesting, did they implement it themselves or is there a library - I mean the graphical component and the financial calculations.
As for the results, it seems that profitability and Sharpe Ratio are not enough - almost no margin for slippages and commissions, if there are any.
the tester for python, liba - there are plenty of different ones
As for everything else - now I drive with different parameters and I lose enthusiasm, the same overfit as in the forest
it's not hard to understand where the trainee is and where the test is. That is, in fact, nothing has changed, catbust did not give advantages.
Later I will try lstm.
Regarding everything else - now I race with different parameters and enthusiasm disappears, the same overfit as the forest
What are the chips and targets?
What are the features and targets?
The increments are normal, the targets are random from training to training, through different steps (like a zigzag with floating parameters)
increments are normal, targeted randomly from training to training, through different steps (like a zigzag with floating parameters)
ok
i had no good results with returns or increments, it was too noisy, i couldn't trade that way((
clearly
It's not so good with returns or increments, it's too noisy, you can't trade that way(((.
If they do, they are short-lived, or the spread is smaller. If they do, they are short-lived, or there is less of a spread.
You can take more order and fewer trades, but there are no normal patterns. If there are, they are short-lived, or less spread
What do you expect...
Of course, you can fight noise in many ways, but it comes out as "axe soup".
Well, what did you want...
It was just interesting to compare the classifiers
I didn't get much from the screenshot.
It was just interesting to compare the classifiers
I did not understand much from the screenshot
classifier - forest, chips - momentum signs(10,20,40,80,160,640,1280,2560,5120) target - direction sign ZZ(10)
there's nothing to compare, it's a lame configuration
classifier - forest, chips - momentum signs(10,20,40,80,160,640,1280,2560,5120) target - direction sign ZZ(10)
Have you tried to montecarrelize the fiches\returns? and add them to the training sample. I.e. make several implementations of the process with different drifts etc. The only thing I haven't done yet
because when we take differences from prices, we lose amounts, fiches get incomplete. Montecarlo can be fixed... probably
https://programmingforfinance.com/2017/11/monte-carlo-simulations-of-future-stock-prices-in-python/
Have you tried to montecarrelize fiches\returns? and add them to the training sample. I.e. make several implementations of the process with different drifts etc. The only thing I haven't done yet
because when we take differences from prices, we lose amounts, fiches get incomplete. Montecarlo can be fixed... probably
https://programmingforfinance.com/2017/11/monte-carlo-simulations-of-future-stock-prices-in-python/
I tried lots of things, IMHO it's a sad business to fix the sign for future increments, at least I never learned how to do anything good with it, it's not about peculiarities of configuration but close to zero predictability, which is completely leveled out by trade costs.
Increments are "micro"-level, like movements of atoms, and it's necessary to focus something less noisy, something like trendiness/flatness, than to filter the usual TS of pullback and impulse ones.