Don't tell me then that TA doesn't work - page 11

New comment
 
bolt:

Is it possible to set the boundaries of the test and optimisation time intervals myself? Have I correctly understood that the results of parameter optimisation are in the tester\golddust file?

Can I replace an in-house Expert Advisor with my own one under the title of in-house expert and how to do it correctly? Thank you.


Only if you optimise it manually. There is no provision for this in the software.
 

I'm sick and tired of all the scum flooding this resource. They'll bury any idea. The creatures lack a respectful tone in their address, but they can't even get off their asses and look at the source code and get into the idea. They could at least write something constructive. All you do is jerk off. You're breaking the main rule of the forum - respect.

It's a total disrespect to start criticizing without understanding. That's what all the bums do, not just here, but in life in general. Idlebodies who cannot give birth to anything useful but are quick to criticize any idea.

Moderators: If a forum user is a freeloader - I do not respect him.

Topikstarter: There are a few research ideas.

  1. Move away from TP and SL and go to the model of signals: BUY, SELL, BUY.
  2. Accordingly, BP of signals will consist of three values: +1, -1, 0.
  3. Optimization on the first interval, we obtained TP1.
  4. At the second interval we have obtained BP2.
  5. Optimization on two BP3 intervals at once.
  6. We need to visualize all BPs.
  7. See the correlation between BP1, BP2, BP3 and (BP1 "crossed" by your proposed methodology with BP2) at different intervals.

Objective, to see and analyse how the fitting component is subtracted from BP.

So far the following is clear, the more BPs are "crossed", the more zeros the new BP will contain.

P.S. The proposed research idea is not a recommendation from "smart" to do so. It is something that I will do, and then the results, concerning this particular branch, constructively and, perhaps, with hot words, discuss here, expressing my research respect.

 
hrenfx:

Topikstarter: there are some research ideas.

  1. Move away from TP and SL, and go to a signals model: BUY, SELL, smoke.
  2. Accordingly, the BP of the signals would consist of three values: +1, -1, 0.

I am trying this idea but the results are not satisfactory yet. I.e. there is a TS without takeoffs and stops that can be well adapted to the history. But it behaves very unstably on forwards with a filter.

The main thing here is profit probability on OOS. If the profit on OOS is obtained only at the N-th try, the probability that we are not dealing with a fitting is approximately equal: 1 / N

Stability in forward testing is most important, the rest can be improved and improved later.

 

I think we should analyze BP (+1, -1, 0) for all TS for the fitting component.

I.e. we take any TS, convert its history into our BP according to the following rule, where BUY net position we write +1, SELL - -1, net zero position - 0.

Write a usable toolkit for investigation of these BPs obtained by optimization at different intervals and "crossings".

It is clear that in SB absolutely all TPs are fitting. That's why the toolkit for SB should always show the same picture. Well this is more of a chit-chat. I will do it and share with you.

 

Came up with a much better idea than (+1, -1, 0). Can be tested with any MM.

Compose BP as volume (sign, as direction) of the net position. Then "cross" BPs according to the rules:

  1. Divergent - crossing signal zero.
  2. Unidirectional - crossing signal at least BP.
 
hrenfx:

Optimisation on two BP3 intervals at once.

It seems to me that this completely destroys the idea of mutual filtering of two signal modules trained on different intervals?

And the departure from the SL and TP, it is correct, and in this case and in general. For simplicity we can even take (+1, -1), and 0 we will get and already at joint work signal modules and opposite signals from them.

Well of course the inputs should be changed, "wretchedness" of the input data in this case can level the pluses of the idea in the final result.

I will try it now too.

 
Figar0:

It seems to me that this completely destroys the idea of mutual filtering of the two signal modules trained at different intervals?

The idea is fully retained. BP3 only needs to be compared for research purposes. Just at least to see how the fitting component looks like.
 
Reshetov:

Yeah. This thread has gone on and on and on.

Why don't you say something?

Rule number eight is already in effect.

I've experienced it myself....

It's sobering!

 

Man, I don't know about anyone, but I got similar results almost immediately. Who doesn't like what's in them???? The lot is permanent...

It's far from OOS, but the lot after it is from August to February 2011

 
Bravo! // that's what I'm saying - it's not that complicated in reality...
1...456789101112131415161718...27
New comment
Reason: