"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 45
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
About initialization: don't you get it?
Initially the grid (weights) is mostly initialized by random.
It would be nice to be able to initialize with zeros, then you can immediately see what the learning algorithm is capable of.
Also, if you don't want to learn the net all over again, we load weights from the file and finish learning.
Initially the grid (weights) is mostly initialized by random.
It would be nice to be able to initialize with zeros, then you can immediately see what the learning algorithm is capable of.
Also, if you don't want to re-learn the net, load weights from a file and finish learning.
At nothing, with zero grid weights most algorithms will not move at all.
Loading grid weights is a standard task, and naturally it will be present, it should be possible both to start from random and from a file of weights (this is by default, not even discussed).
Nothing, from zero grid weights most algorithms won't move at all.
You can immediately see what the learning algorithm is capable of.
from the random and from the scales file(this is the default, not even a discussion).
It's easy to see what the learning algorithm is capable of.
People, on the contrary, are working hard, not sleeping nights to figure out how not to get into the zero hole, and you yourself are right there with your head.
I, too, sometimes I don't sleep at night, but I have such a tough test. Try it.
Otherwise, the grid and the algorithm are not capable of anything.
If you compare it to the human brain: when a baby is born, its neurons are full of zeros. )
I, too, sometimes I don't sleep at night, but I have such a tough test. Try it.
Otherwise, the grid and the algorithm are not capable of anything.
In most cases (except for those single ones when you need 0 weights, which are the cases when the grid is not needed at all, because neurons are zeroed out) training will only slow down and that's all (and not significantly). There is no positive side of 0-weights in initialization.
For the sake of what would "be" you can certainly do, but there is no practical sense in it.
In most cases (except those sporadic cases when 0-weights are needed, which are the cases when the grid is not needed at all, since neurons are nulled) training will only slow down and that's all (and not significantly). There are no positive sides of 0-weights during initialization.
For the sake of "would be" can certainly be done, but there is no practical sense in it.
I will not argue.
The main thing is to be able to load from a file.
In the extreme case (mine), you can stuff the file with zeros.
I'm not going to argue.
The main thing is to be able to load from a file.
In the extreme case (mine), you can stuff the file with zeros.
What kind of mood is this, to change the load weights is two lines of code, people here generally are not going to write code?
The project is called "engine neural network," no one will wipe your snot, will make the engine, and tighten the nuts and change the wheels will be your own.
And no one will write an automaton for you with a [dough] button.
If you compare it to the human brain: when a baby is born, its neurons are full of zeros. )
I was immediately reminded of a bearded anecdote.
I doubt about "zeros" at the moment of birth, besides I think that errors-deflections are present all the timeI was immediately reminded of the bearded joke
I doubt about "zeros" at the moment of birth, besides I think that errors-deflections are present all the time