Indicators: Sampler

New comment
 

Sampler:

The indicator (i_Sampler.mq5) calculates perfect market entry signals, which can be used for neural network training.

It has two buffers:

  • buffer 0 (green line) - analog signal, calculated as a ratio of the positive/negative price change (normalized to range [-1, +1]) during bars_future bars forward;
  • buffer 1 (two-color histogram) discrete signal, the values are: -1(sell), 0(wait), +1(buy).

The discrete signal can be calculated 2 different methods:

  • method 1: the signal appears when analog signal became greater than threshold, specified by porog input parameter;
  • method 2: the signal appears when TP/SL conditions are satisfied.

The e_CheckSampler.mq5 Expert Advisor is created for checking of the indicator values.

Author: her.human

i_Sampler indicator

 
Automated-Trading:

Sampler:

Author: her.human

It is a little unclear what is the ideal inputs for NS.

They (inputs) what? do not contain contradictory patterns? or what other criterion of ideality?

 
Urain:

It is a little unclear what is the ideal inputs for NS.

Do they (inputs) what? Do they not contain contradictory patterns? Or what other criterion of ideality?

Inputs from the point of view of trading.

The indicator shows what should be taught to NS, when learning with a teacher. I.e. what should be at the output of the network.

 
her.human:

Entries from the point of view of trading.

The indicator shows what you should teach the NS, when learning with a teacher. I.e. what should be the output of the network.

Have you done a contradiction check?

This is when the past of an example is similar, but the future is multidirectional.

 
Urain:

Have you done a contradiction check?

It's when the past of the example is similar, but the future is divergent.

"The past of the example is similar" is similar to what?

It depends on the data fed into the network input.

The indicator only calculates the desired output of the network (in several variants), it does not calculate any future or inputs. What to feed to the input is another topic.

For a rollover system (always in the market) - it is better to use 1 signal method or analogue buffer.

For a system with fixed stop loss and take profit - the 2nd method.

I made the indicator to simplify the preparation of training examples.

How the network will identify these examples - it depends on its abilities.

 
her.human:

"the past of the example is similar" is similar to what?

It depends on the data fed to the network input.

The indicator only calculates the desired output of the network (in several variants), it does not calculate any future or inputs. What to feed to the input is another topic.

For a rollover system (always in the market) - it is better to use 1 signal method or analogue buffer.

For a system with fixed stop loss and take profit - the 2nd method.

I made the indicator to simplify the preparation of training examples.

How the network will identify these examples - it depends on its abilities.

Let's make a fool of ourselves, come on.

"The past of an example is similar" is similar to what?

There is a correlation between two examples (for example), an example (training example) is an example because it has a past (those inputs) and a future (those desired outputs).

In this context, the biggest problem is when the same past of two examples has different future, such examples cannot be learnt by the grid in principle. And this is the main problem why nets on vr don't work.

 
Urain:

Let's play dumb, come on.

I didn't mean to. I thought I misunderstood the question.

There is a correlation between two examples (for example), an example (training example) is an example because it has a past (those inputs) and a future (those desired outputs).

In this context, the biggest problem is when the same past of two examples has different future, such examples cannot be learnt by the grid in principle. And this is the main problem why nets on vr don't work.

Now I know I got the question right.

I don't know yet what I will feed to the grid input (I am still searching for optimal solutions). But I already know exactly (this indicator will tell me) what should be at the output.

One of the options to solve your problem (imho):

Take two grids, train one for purchases - the other for sales. Then if both grids give multidirectional signals - ignore the signal.

 
her.human:

I wasn't even going to. I thought I misunderstood the question.

Now I know I got it right.

I don't know yet what I'm going to feed to the input of the network (I'm still searching for optimal solutions). But I already know exactly (this indicator will tell me) what the output should be.

One of the options to solve your problem (imho):

Take two grids, train one for purchases - the other for sales. Then if both grids give multidirectional signals - ignore the signal.

Yeah, well, the idea is clear, it has its place. I think many networkers will be interested.

Thanks for the posted development.

 
It doesn't redraw?
 
DimaD:
It doesn't redraw?
Of course it does. There's no point otherwise.
 
DimaD:
He's not redrawing?

No, he doesn't redraw.

P.S. He only draws.

123
New comment