Machine learning in trading: theory, models, practice and algo-trading - page 324

New comment
 
Maxim Dmitrievsky:


And what does it mean when the results start to go into turbulence during genetic optimization? :) The graph should improve with time

Maybe you should decrease or on the contrary increase the step. I.e. either misses or cannot get out of the pit. Correction, I don't know exactly how this is organized in MT.
 
Yuriy Asaulenko:
Perhaps it is necessary to reduce or on the contrary to increase the step. I.e., either misses, or can not get out of the pit. Correction, I do not know how it is organized in MT.


And it started to occur at different timeframes during runs... It's strange... nevertheless I got 100% for one month on 5-minute trades :) and the number of trades is quite large, almost 200. That is, it is not quite a coin. For this reason I will try to place the real money in the opposite direction.


I also noticed the strange thing, that 1/4 forward is always worse than 1/2, I think it's because of decreasing amount of history for backtest and it retrains not so much, but it's only a guess.


 
Maxim Dmitrievsky:


And on different timeframes during runs began to occur... Mmda, strange... nevertheless 100% for a month on 5 minutes :) and the number of transactions delivers, almost 200. That is not quite a coin

How do you do on M1?
 
Renat Akhtyamov:
How are you doing on M1?

Now running in the cloud, I'll take a screenshot when it's done
 
Maxim Dmitrievsky:


I've got a feeling that I'm not going to use automatic optimizing, but I've nevertheless got 100% for one month on 5-minute trades :)

I probably have about the same -10-16%/month. Only I count not on the depo, but on the volume of the transaction - imho, better reflects reality. It should be recalculated over the shoulder.

By the way, I do not use genetics. And in general I am wary of auto-optimization.

 
Yuriy Asaulenko:

I probably have about the same -10-16%/month. Only I count not on the depo, but on the volume of the transaction - imho, better reflects reality. It should be recalculated over the shoulder.

By the way, I do not use genetics. And in general I am wary of auto-optimization.


What type of NS?
 
Maxim Dmitrievsky:

Also NS? What kind of NS?
Not NS yet. Logic, but it is so complicated that I can no longer cope. Further complication is a direct way to mental hospital.) Which leaves me only Data Mining. There's not much choice.)
 
Yuriy Asaulenko:
Not NS yet. Logic, but already so complicated that I can no longer cope. Further complication is a straight path to madhouse.) This leaves me with Data Mining. There's not much choice.)

That's why I took up NS, because I can't handle tons of code with logic)) like let it all come up by itself
 
Renat Akhtyamov:
How are you doing on M1?


Something similar on M1:

1. it is necessary to adjust the logic for a more high-frequency trading, trailing is not enough here

2. Buy or sell orders prevail on different runs, what may indicate the retraining. 2 weeks of studying is too much for minutes...

 
Maxim Dmitrievsky:


On the minutes something similar, but:

1. we need to adjust the logic for more high-frequency trading, trailing is not enough here.

2. buy or sell orders prevail on different runs, which again may indicate overtraining.

Yeah.

I have never been used to trade in the past, I have never been used to trade in the past.

1...317318319320321322323324325326327328329330331...3625
New comment