Machine learning in trading: theory, models, practice and algo-trading - page 3225

[mytarmailS]

fxsaber #:
Doing this.

What you are doing in points 3 and 4 is some kind of self-made and not very high quality attempt to solve the problem of optimising a noisy target function.

You are trying to find not one set of parameters, but some clump / cluster of close sets of parameters.

I think you should read about optimisation of noisy functions, maybe you will find something useful.

[Andrey Dik]

mytarmailS #:

I think you should read about optimising noisy functions, you might find something useful.

is it something special "noisy functions"?

optimised functions of trading strategies are all noisy.

There are no smooth ones and cannot be smooth because of the discrete nature of the DEM and discrete actions of the TS.

[fxsaber]

mytarmailS #:
What you are doing in points 3 and 4 is a kind of self-made and not very good attempt to solve the problem of optimisation of a noisy target function.

You are trying to find not a single set of parameters, but some cluster / cluster of close sets of parameters.

I think you should read about optimisation of noisy functions, maybe you will find something useful.

I would read something practical: how to find market patterns (from practicing algo-traders).

ZY You participated in this good discussion.

[Renat Akhtyamov]

fxsaber #:

I would read something practical: how to find market patterns (from practicing algo traders).

Try writing an EA.

It looks pretty good already.

but a 45 degree angle indicates an equally likely direction of movement - up or down.

So the probability of buying/selling is the same, 50/50.

I assume that the target in your research should be a line that goes up and down.

but

maybe I'm very wrong

[fxsaber]

Renat Akhtyamov #:

the purpose of your research

is to find a solution so that you can

Forum on trading, automated trading systems and testing trading strategies

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

fxsaber, 2023.08.19 10:36 am

Create a lot of scalping stories. And on them to identify the vulnerabilities of the TS. Now there is stupidly not enough history length for such checks. Therefore, adequate generation is needed.

We have not been able to do it with the means of the MoD so far.

[Maxim Dmitrievsky]

fxsaber #:
is to find a solution so that we can

.

IO tools have not yet been able to do this on the fly.

It is desirable to have more initial information about the TS, otherwise we get a search for unknowns with approximately these data dimensions

[1000][6930269], which is not very fast.

at least average holding time of a trade, or average number of ticks between opening and closing to limit the search area

if some other piece of the past history is analysed before the trade, it should be included as well

and I don't have a supercompukter at hand yet :)

[fxsaber]

Maxim Dmitrievsky #:

It is desirable to have more initial information about the TC, otherwise it turns out to be a search for something unknown with approximately these data dimensions

[1000][6930269]

at least average trade holding time, or average number of ticks

On TesterDashboard screens the blue numbers are: profit in pips, number of trades, PF, average profit per trade.

It's not about the TS. Here we have a pattern in the TSVR. We start generating it, but it is not there. Probably, it is not very good for the MO.

[Maxim Dmitrievsky]

fxsaber #:

On the TesterDashboard screenshots blue numbers: profit in pips, number of trades, PF, average profit per trade.

It's not about the TS. There is a pattern in the CEVR. We start generating it, but it is not there. Probably, it is not very good for the MO.

Your TS is not looking for all patterns that may exist. That's why you have to match it

Otherwise through another approach, but it will be long on ticks and later )

[Maxim Dmitrievsky]

Maxim Dmitrievsky #:

Now there is some interconnectivity within the chains, 100 ticks long

I can make it to a different depth, if it coincides with the average life of TC positions, it should work, in theory.

https://disk.yandex.ru/d/PnU3K-tUgmu-oA

dependence with the length of 1000 ticks

https://disk.yandex.ru/d/6F8FdUGthpnk3A

then I will try to understand the approach and make tests, but I managed to speed up the calculations.

It may not save any memory in sequences, but it is fast :)

And with a length of 5k, on top of that

https://disk.yandex.ru/d/1ypCrzYKk82XdA

[Aleksey Vyazmikin]

fxsaber #:

Maybe we should try to solve the problem of generating new data through approximation?

Take a window and try to describe a numerical series in it with different accuracy, while this approach will allow to save the dynamics of price movement globally, including daily fluctuations.

And, it will be enough to save the history in the form of approximators coefficients.