Machine learning in trading: theory, models, practice and algo-trading - page 3123

New comment
 
Renat Akhtyamov #:

How tiresome to watch it all: the exploration, the puzzles and the system searches....

The logic of constructing a market quotation could not have been complicated in 1970.

What does neural networks have to do with it, if the quotation was launched at that time from the knees, from a paper written with a pencil and counted on the accounts?

So what if it's been 50 years.

The algorithm hasn't changed, I'm telling you as it is, 100%, tested!

Well, in 1970 a man couldn't invent something that a man couldn't understand, he couldn't!

There's a new collection of underwear at the mall, go and see.
 
Maxim Dmitrievsky #:
You can even derive the difference and adjust it through the MOE.
Difference to what? It's the OOS, which is unknown. Everything is fine on the trayne, there is nothing to calculate the difference with.
 
Forester #:
Difference to what? It's the OOS, which is unknown. It's fine on the train, there's nothing to calculate the difference with.
Compare the OOS to the traine for starters. The trayne would be the tritment group and the OOS would be the control group. You can first just look at the shift in the mean of the traits. If there is one, then look at the dynamics of such shifts throughout the history. If it is possible to cure then without taking OOS into account, then good :)

If there are a lot of traits, it's quite a creative challenge. I haven't got through it all yet.

The task essentially boils down to how to fix bias. This is a target task after I've learnt how to put numbers into the model. If it can't be fixed in any way, it's a lousy job, of course. But that's no reason to give up (I guess) 😀
 
Maxim Dmitrievsky #:
You can even derive the difference and adjust it through the MOE.

The difference of what?

it is clear, as you say, the meta parameter of a series is its mathematical model, and the parameters of the model are the parameters of the series, but the models are different and sometimes one has the parameters, the other does not have them or the behaviour of the model from the parameters is different. And to compare the results of the model in the form of TC ... I don't think it's correct.

Probably there can be a dependence of correlation of some parameters of a series on its behaviour. It's raw, of course...

What do you think about modelling trade negotiations?

Machine learning, specifically thekernel method, was used by Renaissance Technologies in the 1980s,

Machine learning, specifically the kernel method ,

What's that in today's parlance?

 
Valeriy Yastremskiy #:

The difference of what?

As you say, the meta parameter of a series is its mathematical model, and the parameters of the model are the parameters of the series, but the models are different and sometimes one has the parameters, the other does not have them or the behaviour of the model from the parameters is different. And to compare the results of the model in the form of TC ... I don't think it's correct.

Probably there can be a dependence of correlation of some parameters of a series on its behaviour. Raw, of course...

What do you think of modelling trade negotiations?

Machine learning, specifically the kernel method, was used by Renaissance Technologies in the 1980s,

Machine learning, specifically the kernel method ,

What is it in today's parlance?

Depends what kind of kernel 😀 polynomial or radial basis or something. It's fine in today's parlance. The model is shallow (if in regression or support vector method), but it is simple and interpretable.

The difference between distributions and the model's response to them. It seems to be very obvious. It remains to figure out how to level it out.
 
Maxim Dmitrievsky #:
Compare the OOS to the traine for starters. Train will be the tritment group and OOS will be the control group. You can first just look at the mean trait shifts. If there is one, then look at the dynamics of such shifts throughout the history. If it is possible to cure then without taking OOS into account, then good :)
.

If there are a lot of traits, it's quite a creative challenge. Haven't got it all down yet.

The task essentially boils down to how to fix bias. This is a target task after I have learnt how to put numbers into the model. If it can't be fixed in any way, it's a lousy job, of course. But that's no reason to quit (I guess) 😀
The model on sell starts to sag when the global (just 1-1.5 years) trend is up. It finds an opportunity to make money on the trade, but on the OOS it goes into drawdown.
Maybe the first option with buy|sell selection by one model will be better. But if she adjusts to the global trend, she will drain at the moments of trend change. And probably will trade in one direction for years.
 
Forester #:
The model on sell starts to sag when the global (just 1-1.5 years) trend is up. It finds an opportunity to make money on the trade, but on the OOS it goes into drawdown.
Perhaps the first variant with buy|sell selection by one model will be better. But if she adjusts to the global trend, she will drain at the moments of trend change. And probably will trade in one direction for years.
The model is biased. So we need to force it to learn without such a bias. But first we need to find the bias coefficients, let's say it is a slope or a free term (intercept), as in regression. What if we make it train in such a way that this term does not vary by traine and OOS. Basically quoting books on kozul.

In catbusta and other models, you can assign weights to labels during training. For example, the offset is output, then converted to weights and the model is trained with correction factors already on the traine. This is one of the ways.
 
Maxim Dmitrievsky #:
There's a new collection of underwear at the mall. Go and see.

I remember your 0.1% risk on your deposit.

Don't bother with advice.

It's nothing.

I trade at leverage 2000 with 95% risk and pay attention to advice, experience and so on, only from experienced and successful like me.

 
Maxim Dmitrievsky #:
Goodbye, blabbermouth. Go watch football.

That's pretty good.)

write poems and books.

Go for it.

It's yours, and it's probably more profitable.

 
Renat Akhtyamov #:

wha

to write poetry and books

get into it.

It's your thing and it's probably more profitable

You can go on YouTube if you're so dumb.
1...311631173118311931203121312231233124312531263127312831293130...3699
New comment