Machine learning in trading: theory, models, practice and algo-trading - page 2962
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
The main thing is to believe in it and repeat it more often, or vice versa.
I realise it's frustrating when you've sort of done something and you're told it was pointless in the first place :)
Did I ever brag about the cuts?
I'm saying that there are no limitations in training through FF, but there are limitations in training with targeting....
Did I ever brag about the cuts?
I'm saying that there are no limitations in training with FF, but there are limitations in training with targeting.
Did I ever brag about the cuts?
I'm saying that there are no restrictions in training through the FF, but there are restrictions in training with targeting.
You're the one who started arguing with me and Fomenko picked up on it.)
Have you changed your mind?
I'm just trying to move in a practical direction, to make a grail, for example.
You just need to look at the concept of FF from a different angle. minimising error is also FF.
In fact, the choice of FF is no less important than the choice of training data. The FF should take into account what the model should do and what it should not do, i.e. maximise the necessary TC actions and minimise the unnecessary TC actions.
You were the first to argue with me, and Fomenko picked up on it...)
Change your mind, though?
So I was arguing with you that FFs are no better than targets?
Where did that come from?
I smell bullshit.
So I'm arguing with you that FFs aren't better than targets?
Where was that?
I smell bullshit.
I originally said that all learning is maximising the FF.
you and Fomenko argued that such an approach would lead to overtraining - at the same time, I said that the FF is not properly designed if overtraining occurs.
I don't know what you're sensing.)
I see that you have given up the habit of "poking" strangers, that's already good.
I originally said that any learning is maximising FF.
you and Fomenko argued that such an approach would lead to overtraining - at the same time, I said that the FF is not properly designed if overtraining occurs.
I don't know what you're sensing.))
I see that you have given up the habit of "poking" strangers, that's already good.
I'm sure then a long time ago, the conversation was a little different, and now in general quite different and different ....
Yes, after New Year's Eve, I've decided to address everyone as "you".
You just need to look at the concept of FF from a different angle. minimising error is also FF.
In fact, the choice of FF is no less important than the choice of training data. The FF should take into account what the model should do and what it should not do, i.e. maximise the necessary TC actions and minimise the unnecessary TC actions.
The only option I see is to optimise it off-train, on test, and train the MO on-train. That still makes some sense in terms of pulling something common from different historical periods
There is a lot of room for imagination in how to use FF, it's a pity that there is no single correct recipe for FF preparation, although there are some recommendations.
For example, optimising on balance. The TS showed 90% of profit in one trade, and the rest of the hundred trades are about zero. is this a good FF? maybe it is good, but not for this strategy. and maybe the strategy is so bad since there are such variants of results in optimisation.
Thus, the FF must take into account everything that is required from the model and the problem is reduced to maximising the global optimum. and another conclusion follows - the model and FF cannot be by themselves, a good model can be spoiled by an unsuitable FF and vice versa is also true. although it makes no sense to talk about FF separately from the model in general.