Build Neural Network Indicator in MT4 using Neuroshell

Hi AT,

Please attach your data file including Date OHLCV and all indicators value that you are using, if you have them in an NST chart please attach cht file with saved data. I will try to model it.
 
Last edited:
Hi Arryex and Tovim!

@tovim: yes maybe, but when I use a model only unsing MACD I get (better) results,
only one indi used... (of course different software, different problems... ;-) )

@Arryex: Thanks a lot... data attached.
Maybe a logic failure of the model?!

Maybe my CPU speed is the problem... ;-))

Bye, AT
 

Attachments

  • eurusd_JMANS55_5_26_ADX_CCI.zip
    146.3 KB · Views: 232
Hi AT,

First, I am surprise when I call your saved model it doesn't show a proper output, I found that your csv data format is different with mine (normal csv format).

What actually software that you use to create csv file? otherwise we have different PC formatting,i.e. "," delimited is converted to ";" delimited on your PC and "." become ",".

i will try to change it first.
 

Attachments

  • data file.png
    data file.png
    107.9 KB · Views: 398
  • CH output.png
    CH output.png
    105.8 KB · Views: 366
I found few possible models from your data as shown here, it may not be good model since I do not make exhaustively just continue from your original setting.
 

Attachments

  • moda2-opt.png
    moda2-opt.png
    127.2 KB · Views: 351
  • moda2-oos.png
    moda2-oos.png
    129.6 KB · Views: 323
  • moda2-all.png
    moda2-all.png
    122.6 KB · Views: 400
As shown here while modelling, I do not select the last model, since the last model might be overfit. FYI, I am not using a supercomputer, just a normal PC, it takes only few minutes. BUT "the previous performance should not guarantee future performance".
 

Attachments

  • moda-processing.png
    moda-processing.png
    127.7 KB · Views: 335
Hi Arryex!

Thanks for your help!

My data format looks like the following (attached) when I open it with an editor.
Of course it's different - I think that's results of US and EU? hmmm.
I also have this when I open a (US-)Excel file... "," = "."
Maybe I can change this setting...

Anyway in CH - when I look all data - the are shown.
Also when I save md with CH, look ok...

Bye, AT
 

Attachments

  • data_editor_MD_file.JPG
    data_editor_MD_file.JPG
    40.7 KB · Views: 229
  • data_editor_CSV.JPG
    data_editor_CSV.JPG
    66.1 KB · Views: 220
Last edited:
Hi arryex!

Thanks a lot for your calulation!! I have no such results... far away from it!!

What do you mean no "supercomputer"?
How many generations have you made?

I had - 20000 or so... (2,8GHZ CPU sinlge - CPU time about 20hours)

Bye, AT
 
Last edited:
Hi Arryex!

Please, see the pictures attached... no good model...
Strange thing for me -> about 20.000 generations...
(training time was about 14hours...)
But, model info say, only 4500?! generations and strange optimization time.
(maybe this is the time/info CH needed for the choosen model - anyway)

Bye, AT
 

Attachments

  • CH_trash_model_EU_H1_2_info.JPG
    CH_trash_model_EU_H1_2_info.JPG
    192 KB · Views: 364
  • CH_trash_model_EU_H1_2_settings.JPG
    CH_trash_model_EU_H1_2_settings.JPG
    119.3 KB · Views: 335
  • CH_trash_model_EU_H1_2.JPG
    CH_trash_model_EU_H1_2.JPG
    214.5 KB · Views: 1,142
Hi Arryex!

I used the formular editor and tried to "rebuilt" your results/model...
(thanks for this!)
(not what I want to do in future ;-)) please, don't get me wrong, very very nice results/models from you, but I of course want do it by my own)
and got the following results ?!
I think, I did something wrong.... any ideas...

Bye, AT
 

Attachments

  • CH_trash_model_EU_H1_no_rebuilt_3.JPG
    CH_trash_model_EU_H1_no_rebuilt_3.JPG
    236.2 KB · Views: 326
  • CH_trash_model_EU_H1_no_rebuilt_2.JPG
    CH_trash_model_EU_H1_no_rebuilt_2.JPG
    235.2 KB · Views: 349
  • CH_trash_model_EU_H1_no_rebuilt.JPG
    CH_trash_model_EU_H1_no_rebuilt.JPG
    241.9 KB · Views: 377
Hi Arryex!

Thanks for your help!

My data format looks like the following (attached) when I open it with an editor.
Of course it's different - I think that's results of US and EU? hmmm.
I also have this when I open a (US-)Excel file... "," = "."
Maybe I can change this setting...

Anyway in CH - when I look all data - the are shown.
Also when I save md with CH, look ok...

Bye, AT

I think we have different format setting on windows (I use USA formatting by default)
 
Hi Arryex!

I used the formular editor and tried to "rebuilt" your results/model...
(thanks for this!)
(not what I want to do in future ;-)) please, don't get me wrong, very very nice results/models from you, but I of course want do it by my own)
and got the following results ?!
I think, I did something wrong.... any ideas...

Bye, AT

Hi AT,

When you do the optimization, you set the optimization and out of sample range right? at the time CH doing the optimization, select the equity instead of objective. Then you can see how the equity on both optimization and oos. You select which model that giving good result on both sides (optimization and oos).

That what so far I did, I never let CH running so many generation if the result is not good, I will change the input formula if necessary.

What I means as supercomputer, I do not have GPU, I just use 8 core CPU.

Hope this will help you.
iinvestsaham
 
Hi arryex!

I think i got your points... One thing I don't understand is the point with the
rebuilt formular from you, which doesnt work... Any ideas?

Thanks and bye, AT
 
Hi At,

The case we face here is a common problem of prediction with AI, a model tends to overfit during optimization period, more time used during optimization more the model to overfit and worst during oos.

I use the method avaible in most software, set the optimization, paper trade and oos, in some software they call is as optimization, test and production range. An ideal model should perform in those data range and lucky if it works on the real data.

You should find how to perform such modeling, CH only provide optimization and oos while NST has papertrade option.

Cheer
iinvestsaham
 
Hi Arryex!

I understand the problem with overfitting. training, papertrading, OOS, -> real trading, but my problem actualy is, that I'm not able, to "rebuilt" or let me say setup, the model you have in your sample posted. I use the same data and the same settings (input via "formular editor") so for my understanding I should get the same results for the same time/date?

Bye, AT
 
I do not know why it happen, FYI the model edited on editor could not be reoptimized. I will resend you the data which I modified from yours later, I am still on the way right now
 
Hi Arryex!

This would be great, so I can compare it and maybe find my fault.
Formular editor -> no optimization, ok.
I haven't done any optimization, I just want to rebuild it 1:1.
But no success! hmmm.

Thank you!

Bye, AT
 
Here the data that used, since my excel read the csv in "," delimited then I change all ";" for column separation. See the display comparison your original data with mine, your original one could not be shown in my CH.
In addition, you my ask to ward support center related with your concern, they may give you an advice.
 

Attachments

  • eurusd_JMANS55_5_26_ADX_CCImod.zip
    142.8 KB · Views: 202
  • comparison in excel.png
    comparison in excel.png
    133.5 KB · Views: 408
Hi Arryex!

Thanks for your info...
I think data input is not the problem... see the attached files...
(I also have to convert sample files from ward if I want to test/use it)
Also attched the screenshots of the whole setup...
Very very strange... if I now input your formular I got bad results...
No matter wether I use "scale inputs" or not..
Do you see any faults in setup?
Otherwise, I'm going to have to contact ward support...

Thanks and bye, AT
 

Attachments

  • CH_trash_model_EU_H1_opti.JPG
    CH_trash_model_EU_H1_opti.JPG
    67.9 KB · Views: 325
  • CH_trash_model_EU_H1_range.JPG
    CH_trash_model_EU_H1_range.JPG
    110.2 KB · Views: 370
  • CH_trash_model_EU_H1_input.JPG
    CH_trash_model_EU_H1_input.JPG
    54.5 KB · Views: 314
  • CH_trash_model_EU_H1_formula.JPG
    CH_trash_model_EU_H1_formula.JPG
    80.1 KB · Views: 327
  • CH_trash_model_EU_H1_datagrid.JPG
    CH_trash_model_EU_H1_datagrid.JPG
    211.9 KB · Views: 351
  • CH_trash_model_EU_H1_2_settings.JPG
    CH_trash_model_EU_H1_2_settings.JPG
    119.3 KB · Views: 337
Last edited:
here the results of training...
 

Attachments

  • CH_trash_model_EU_H1_equity.JPG
    CH_trash_model_EU_H1_equity.JPG
    144.3 KB · Views: 321
  • CH_trash_model_EU_H1_progression.JPG
    CH_trash_model_EU_H1_progression.JPG
    127.8 KB · Views: 305
  • CH_trash_model_EU_H1_bad_model.JPG
    CH_trash_model_EU_H1_bad_model.JPG
    179.2 KB · Views: 335
Last edited:
Top