Machine learning in trading: theory, models, practice and algo-trading - page 1821

[Aleksey Nikolayev]

Mistakes of the first and second kind? No, we haven't.

[Mihail Marchukajtes]

The news about the possibility of playing KC without installing it from the browser was published on Mile. Now there will be something to do in between signals. IICHUUUUUUU!!!!

Here it is

https://hi-tech.mail.ru/news/cs_1_6_v_brausere/?frommail=1

[Forester]

Maxim Dmitrievsky:

multistep sampling googled. Choice of probability is based on another probability, which is based on another...etc. I don't think there are any other options.

I.e., for example, then trades will open often, then rarely, then not at all. And the time of holding trades will vary greatly.

it may be related to volatility, no one knows how to do it.

Can you give me the link?

[Maxim Dmitrievsky]

elibrarius:

Can I get the link?

multi stage sampling

The signal is very weak (if you can call it that) in quotes, I'm really shocked. If a bit more regularity were added to quotes it would be Ok )

[ipsec]

Maxim Dmitrievsky :

Eu acho que o reforço profundo é usado apenas para tarefas multivariadas, não para 1-5 dimensões, como mercados financeiros. Assim, você pode tentar métodos mais simples e mais rápidos, como REFORÇAR com rede neural ou aproximador linear.

Hi Maxin,

Eu put only six features of last 3 ticks

MA1(SMO) and MA5(SMO) with standardization.

With these features I put to train only if price of last MA1 > bollinger bands upper or MA1 < bollinger bands lower to reduce my time of training.

After training one day in 2018 (take 24hours of training to get lower epsilon) my model is ok to back testing.

I tested in Jan 2019 and getting good results.

So I think the number is dimensions and features is not very relevant if the feature is good enough.

Best regards,

Fernando

[ipsec]

ipsec :
Olá Maxin,

Coloque apenas alguns recursos dos últimos 3 ticks

MA1 (SMO) and MA5 (SMO) with padronização.

Com esses recursos, eu treino apenas o preço das último bandas de bollinger MA1> superiores ou MA1 <bandas de bollinger inferiores, para reduzir o tempo de treinamento.

Depois de treinar um dia em 2018 (faça 24 horas de treinamento para obter o epsilon mais baixo), meu modelo está bem em teste novamente.

Teste em janeiro de 2019 e obtenha bons resultados.

Então, acho que o número de demissões e os recursos não são muito relevantes se o recurso for bom ou suficiente.

Cumprimentos,

Fernando Fernando.

[Forester]

Maxim Dmitrievsky:

multi stage sampling

But in general the signal is very weak (if you can call it that) in quotes, I'm really shocked. If they added more regularity to the quotes it would be Ok)

Do you mean that price quotes are almost random fluctuations?

[Forester]

Maxim Dmitrievsky:

multi stage sampling

But in general the signal is very weak (if you can call it that) in quotes, I'm really shocked. If they added more regularity to the quotes it would be Ok)

I read the first few search results for multi stage sampling. There is only about the cost savings in the research, to survey not 300 million people, but 1000, randomly selecting 10 states, 10 districts in them, 10 people in them, and only them to survey.
We have all the quotes for many years in our tasks. There is no cost to get them.
Apparently you found something else of interest? Not what is in the first search results.

[Mihail Marchukajtes]

So who wants a headshot from around here? Come on in, don't be shy. There you can tell me everything with a gun :-)

https://cs-online.club/ru/servers

[tamagucci]

Mihail Marchukajtes:

So who wants a headshot from around here? Come on in, don't be shy. There you can tell me everything with a gun :-)

https://cs-online.club/ru/servers

On the slipper terra (Glock) on the yyy try) With birst. Years 7 years I have not played, but the skill never lost).

P.S. You can just on the guns. The most fun. Deagle does not count ;)

Train there while.