"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 99

New comment
 
Maxim Dmitrievsky:

I'll look at it later, but I'm not happy with the picture.

For example, for a simple layman, earning on the eurodoll. is unrealistic in principle.

And, by the way, none of the "smart guys" present here can not prove the contrary :) that's all the remains of the 4th forum... ugh. And yet the NS libs were going to write.

Once again, knowing firsthand some participants in this thread and their absolute inability to work in a team, this thread was originally doomed to an ignominious failure.

 
Alexander_K:

Once again, knowing firsthand some of the participants in this thread and their utter inability to work as a team, this thread was doomed to an ignominious failure from the outset.

By the way, it's really more fun with ticks. But there is already a spread there.


 
Maxim Dmitrievsky:

I'll look at it later, but I'm not happy with the picture.

For example, for a simple layman, earning on the Eurodoll. is unrealistic in principle.

And, by the way, none of the "smart guys" present here can not prove the contrary :) that's all the remains of the 4th forum... ugh. And the NS libs were going to write.

The larger the number of participants, the more complex the structure and the more the tool looks like an SB. But it is still not an SB, it is a process with memory. A non-network is unlikely to work, at least by training it on some static data and inputting price values from time.
 
Maxim Dmitrievsky:

By the way, it's really more fun with ticks. But there's already a spread there.


On ticks this is the case for a simple reason: this pattern does not allow earning because of the size of commissions. Everywhere where the size of commissions allows profit, there will be no stable patterns and the more participants, the faster the patterns will break down, and the Eurodollar has the largest number of participants of all markets.
 
Maxim Romanov:
The larger the number of participants, the more complex the structure and the more the instrument resembles SB. But it is still not a SB, it is a process with memory. You can hardly make money with a non-network, at least by training it on some static data and inputting price values from time.

Well, if the pieces of ... floating in the lake are considered "memory", then I guess

if you take the price clause, then there's no memory at all. There are no cycles there, they are just tails, outliers - they have nothing to do with memory. Otherwise there would be a significant difference from SB in terms of entropy, which is not observed. (only bitcoin and some other instruments have it)

 
Maxim Dmitrievsky:

Well, if the pieces of ... floating in the lake are considered "memory", then I guess

if you take the price clause, then there's no memory at all. There are no cycles, they are just tails, outliers - they have nothing to do with memory.

All the more so if you take the clowes! Time sampling itself introduces a large random component. If we discretise music at random intervals, we can't understand anything either.
 
Maxim Romanov:
Especially if you take the cloze! Time sampling in itself introduces a large random component. If we discretise music at random intervals, we can't understand anything either.

That's why Alexander is right about everything :)

 
Maxim Dmitrievsky:

Well, if the pieces of ... floating in the lake are considered "memory", then I guess

if you take the price clause, then there's no memory at all. There are no cycles there, they are just tails, outliers - they have nothing to do with memory. Otherwise there would be a significant difference from SB in terms of entropy, which is not observed. (only bitcoin and some other instruments have it)

Bitcoin will also eventually become so complex that it will seem random.
 
Maxim Romanov:
Bitcoin will also eventually become more complex to the point where it seems random.

yes, but it's interesting... you can get away from the classic tools and look for less effective ones

 
Maxim Dmitrievsky:

yes, but it's interesting... you can move away from classic tools and look for less efficient ones.


That's what I'm talking about: I sliced bitcoin into blocks, at the top, it's 440 blocks, but from 2010 to 2015, you can see some structure. Then I sliced the bottom also into 440 blocks, but from 2010 to 1.06.2019. The same structure is visible, with all the checkpoints, but more complicated. In both cases the growth was around 120 blocks and then a drop of 320


If only to teach the non-network, to recover the data somehow... Or it will have to feed all data of all markets at all timeframes to learn it completely. In general, the data should be prepared somehow.

Or the second option: We need to use very many separate algorithms, with different sizes of analysis windows and varying degrees of learning. Create a population of non-ironets that compete with each other, if need be to become more complex or simpler, as they evolve and trade.
1...9293949596979899100
New comment