Machine learning in trading: theory, models, practice and algo-trading - page 1065

New comment
 
Maxim Dmitrievsky:

Gmdh logic - it will be very helpful, then I can change it for my library

Can you please provide me the formula of GDMH which you are trying to convert to MQL5? I mean there are many varieties and which one of you are trying and implement and how you are trying to implement?

Also, you mentioned " After each deal, update policies, when close trade-update reward (TD, temporal difference RL)

Does it happen in LIVE trading as well?

Or happens only during testing?
 
FxTrader562:

Can you please provide me the formula of GDMH which you are trying to convert to MQL5 ? I mean there are many varieties and which one you are trying and how you are trying to implement?

Also, you mentioned"After each deal he update policies, when close trade - update reward (TD, temporal difference RL)".

Does it happen in LIVE trading also or does it happen only during testing in bactester?

I think we must to use polynomial for all features combinations, for example first line: (x - feature value) x0+x1, x0+x2, ...., x2+x3.... xn+xm where x0,xn are different predictors values

second line x0^2+x1^2...

third line x0^3+x1^3...

with selection of best n results on previous lines ("levels" better). But Im not sure how it works, just learning

... only during testing

 

levels....

levels work....

I'll finish it up a bit more, and then you can hang a neuron as a filter on top of it )

 
Maxim Dmitrievsky:

I think we must use polynomial for all features, for example, first line: (x-feature value) x0 + x1, x0 + x2, ...., x2 + x3 .... xn + xm

second line x0 ^ 2 + x1 ^ 2 ...

third line x0 ^ 3 + x1 ^ 3 ...

with selection of best results

... only during testing

You need to take 2 double arrays...one for weights and one for indicator handles

then, run a for loop for summation of multiplication of weights and indicator handles

Then, another nested for loop for number of features to add all the above summations...I mean from 1 to m where m=number of features

Anyway, I will try to add it to code and let you know when I am done...

Also, I will request you again to somehow make the optimization happen during trading also...otherwise, it will be useless to just run one test and start trading afterwards.I think there is another Mt4 EA which has used R and FANN library, but it saves the optimized data at the end of everyday and hence, I think you can implement the same. I will try to do this also. It will be really simple if we can catch the same values during LIVE trading also and same can be saved to the file.

Also, is it not possible to implement a similar method used in your previous article of direct one click testing instead of changing "Learn" from false to true and again true to false Because I have already my own software to click the "Start" button once in everyday, but probably it can't change the values?

 
FxTrader562:

Also, I will request you again to somehow make the optimization happen during trading also...otherwise, it will be useless to just run one test and start trading afterwards.I think there is another Mt4 EA which has used R and FANN library, but it saves the optimized data at the end of everyday and hence, I think you can implement the same. I will try to do this also. It will be really simple if we can catch the same values during LIVE trading also and same can be saved to the file.

for this we need a virtual tester

anyway I think good features/transformations its a most important thing, and must be done previously

 
Maxim Dmitrievsky:

for this we need a virtual tester

anyway I think good features/transformations its a most important thing, and must be done previously

Yes, I agree.

Is it not possible to implement a similar method used in your previous article of direct one click testing instead of changing "Learn" from false to true and again true to false so that it will automatically save the optimized results Because I have already my own software to click the "Start" button once in everyday, but probably it can't change the values from true to false and false to true.

 
FxTrader562:

Yes, I agree.

Is it not possible to implement a similar method used in your previous article of direct one click testing instead of changing "Learn" from false to true and again true to false so that it will automatically save the optimized results .Because I have already my own software to click the "Start" button once in everyday, but probably it can't change the values from true to false and false to true.

nope, but can use 2 terminals easily - 1 for learning, 1 for trading, because it save models in common folder

but also need to reload EA on trading terminal, or add additional file for checkin if new learning process was

or just to use 1 terminal and learn when you want, EA on chart can update models after it, after some modifications

 
Maxim Dmitrievsky:

nope, but can use 2 terminals easily - 1 for learning, 1 for trading, because it save models in common folder

but also need to reload EA on trading terminal, or add additional file for checkin if new learning process was

or just to use 1 terminal and learn when you want, EA on chart can update models after it, after some modifications

Ok, I think I got it.

1.First, One terminal same EA with "Learn" as true. For first run.

2.Second, second terminal same EA with "Learn" as false. For second run.

Third, third terminal same EA attached to Mt5 chart for trading. Restart this terminal after step 1 and 2 are over.

So repeating this 3 steps everyday will automatically save the optimised results everyday.right?

 
They wrote here about applying volumes to the RTS Index futures. And if you think about it? The RTS Index is a synthetic instrument. What kind of analysis can we talk about? In general, its trading is doubtful. It is the average temperature in the hospital. And the futures is a derivative instrument. Futures repeat the movement of the basic asset, regardless of the traded volumes. So, using the volumes on the futures is just guessing by coffee grounds. It might be applicable to the stocks. There, their movement may depend on them.
 
FxTrader562:

Ok, I think I got it.

1.First, One terminal same EA with "Learn" as true. For first run.

2.Second, second terminal same EA with "Learn" as false. For second run.

Third, third terminal same EA attached to Mt5 chart for trading. Restart this terminal after step 1 and 2 are over.

So repeating this 3 steps everyday will automatically save the optimised results everyday. right?

You dont need 2nd run, its just for check how EA lrarns

1 EA on chart, and learn in the same terminal in tester

1...105810591060106110621063106410651066106710681069107010711072...3700
New comment