Machine learning in trading: theory, models, practice and algo-trading - page 2785

 
Maxim Dmitrievsky #:

TGAN and other GANs, autoencoders, kernel density estimation, copulas

TGAN has not been tried, the others are worse than GMM

Maybe there are new time series GANs available.

*GMM does not converge well on large samples, you need to use not very large ones.

Yes, there are a lot of options, and at the beginning they were even meaningful))) Well, as if a complete search is also a meaningful option, the main thing is to have enough power)))))

 
Uladzimir Izerski #:

Especially for this thread, now for half an hour, using only OHLC, roughly sketched an arrow indicator.

This is the first preview without filters and no other tricks only OHLC. This algorithm will work on any TF.

Like it or not, but no R-key and Python will not help if you do not understand the depth and meaning of financial charts. Sorry for the rudeness, of course.


It doesn't show anything useful.

It's just an ordinary pips indicator.

That's the kind of thing you get scammed out of in a heartbeat.

---

you know, they got one more bar, you can't see it.

They'll give you an up arrow, you'll bounce.

and they'll move it down.

and ala hooey, you're like.

you catch the stop.

and history repeats itself, repeatedly.

turkeys don't work, you should have learnt that by now, you're not a boy.

;)

 

they gave you vectors, wrote an article, and treated you like a trinket....

price movement is a set of parameters that form a vector.

 
Renat Akhtyamov #:

He doesn't show anything good.

pipsqueak

This kind of scam is a one-two punch

---

you know, they've got one bar more, you can't see it.

They give you an up arrow, you biked.

and they'll move down.

and ala hooey, you're gonna be, like.

you catch a stop.

and history repeats itself, repeatedly.

turkeys don't work, you should know that by now, you're not a boy.

;)

They work. They work. You just need to approach the price from the right side.

Now I know for sure that only prices and nothing else should be submitted for the MoD.

 
Maxim Dmitrievsky #:

TGAN and other GANs, autoencoders, kernel density estimation, copulas

TGAN has not been tried, the others are worse than GMM

Maybe there are new time series GANs available.

*GMM does not converge well on large samples, you need to use not very large ones.

How big? 10000 features with a row of 100 each. is that a lot?

 
Evgeni Gavrilovi #:

How big? 10,000 traits with a row of 100 each. is that a lot?

On the edge, it could be a lot. Without PCA, it may not start. Check after sampling and training on this data, how well the original sample is predicted. I don't remember the reason, it seems that EM algorithm doesn't converge when the number of Gaussians is large, and when the number of Gaussians is smaller, it will be soapy at such dimensions

Well, it's like logit regression in the ME world. Good, but not always scalable. And moving to GANs is difficult and it is not clear what architecture to take. The ones for tabular data are definitely worse on time series, and all sorts of ones for time series I haven't tried or forgot
.

 
elibrarius #:
Why do you use your hands? You have a robot. Or do you want some adrenaline from gambling?
I gave up trading with my hands after I lost a few of them.
Or do you want some adrenaline from gambling? I sometimes feel like it too, but experience tells me I shouldn't...
Hello how are your trading bots doing?
 

Transient Response from State Space Representation

Markov Transition (Animated) Plots

only

в Марковских случ. процессах поведение зависит только от значений, принятых системой в наст. момент.

И

in Markov chains the transition of processes to other states occurs by jumps under the influence of random factors...

- timing is not random, even in Taylor series...


Methods of Determing the Transient Response
  • Erik Cheever, Swarthmore College
  • lpsa.swarthmore.edu
csv
 
JeeyCi #:

Transient Response from State Space Representation

Markov Transition (Animated) Plots

only

- There is no such thing as random, even in Taylor's ranks....


In Markov process there is no dependence on the value at the moment.

In general, I understand it as a function of noise. And I understand noise as a process of many factors that cannot be controlled due to their multiplicity. I.e. if there are few factors, the process is controlled, but after a certain number of factors collisions and probabilistic results of summing up the factors begin to appear. Besides, factors can and do have connections and feedbacks. But the Markov process has no such links.

 
Valeriy Yastremskiy #:

In a Markov process, there is no dependence on the value at the moment.

What do you mean, where you are in the transition matrix, you go from there.

Reason: