Machine learning for robots - page 11

New comment
 
Alexander_K2:

Not by price history, but by increments - they form the price (integral of all increments is actually the price from the starting point).

Fortunately, for neural network experts, the 1st condition for Kolmogorov's forecasting (expectation =0) for such BP is satisfied.

The 2nd condition - stationarity - is not satisfied.

I propose to input to the NS, besides the increments themselves, their moments: variance, skewness, kurtosis... and the autocorrelation coefficient. The NS is simply obliged to find regularities in this junk.

I agree, but it is desirable to take data not from a pile of filtered ticks, but from available OHLC bars, which in principle contain also the characteristics you listed.

You need a specific formula, here are some variants I took directly from the training script:

#define  CALC_BAR_1(open,high,low,close) (close-open)
#define  CALC_BAR_2(open,high,low,close) (close-open)/open
#define  CALC_BAR_3(open,high,low,close) (close-open)/(high-low)
#define  CALC_BAR_4(open,high,low,close) (((high-open)-(low-close))/(high-low)
#define  CALC_BAR_5(open,high,low,close) ((open-low)+(high-open)*1000+(close-open)*1000000)/1000000

You can include MQL functions or indicators in the formula if needed:

#define  CALC_IND_1(n) ((iFractals(NULL,PERIOD_M1,MODE_UPPER,n)==iHigh(NULL,PERIOD_M1,n)?1:0)+(iFractals(NULL,PERIOD_M1,MODE_LOWER,n)==iLow(NULL,PERIOD_M1,n)?-1:0))
#define  CALC_IND_2(n) ((iMA(NULL,0,9,0,MODE_EMA,PRICE_CLOSE,n+1)-iMA(NULL,0,14,0,MODE_EMA,PRICE_CLOSE,n+1))>0.0001?1:((iMA(NULL,0,9,0,MODE_EMA,PRICE_CLOSE,n+1)-iMA(NULL,0,14,0,MODE_EMA,PRICE_CLOSE,n+1))<-0.0001?-1:0))

offer your...

 
Ivan Negreshniy:

I agree, but the data should not be taken from a pile of filtered ticks, but from available OHLC bars, which in principle contain some of the characteristics you listed.

You need a specific formula, here are some variants I took directly from the training script:

If necessary, MQL functions or indicators can be included in formulas:

Offer your own...

Of course, this one has to be analysed:

#define CALC_BAR_1(open,high,low,close) (close-open)

And although I work with tick data, I think there should be the following pattern on minute data as well:

* when, at a certain sample size, the modulus of the Pearson asymmetry coefficient for the incremental distribution >= a certain constant, then price is bound to start a "comeback move" (in about 85% of cases), until the asymmetry becomes =0.

I.e. there is already a pattern on one asymmetry. If other points are added, I think it would be even better.

However, this kind of research could take years and I'm getting pretty bored with it...

I think neural networks would detect these patterns faster.

 
Alexander_K2:

Of course, this is the thing to analyse:

#define CALC_BAR_1(open,high,low,close) (close-open)

And although I work with tick data, I think the following regularity must be present at the minute data as well:

* when, at a certain sample size, the modulus of the Pearson asymmetry coefficient for the incremental distribution >= a certain constant, then price is bound to start a "comeback move" (in about 85% of cases), until the asymmetry becomes =0.

I.e. there is already a pattern on one asymmetry. If other points are added, I think it would be even better.

However, this kind of research could take years and I'm getting pretty bored with it...

I think neural networks will detect these patterns faster.

So, roughly speaking - the signed sum of the bodies of black and white M1 candles tends to zero?

If I'm wrong, sorry for my ignorance, and if so, a single adder variable (close-open) instead of a neural network would suffice, but personally I think this is a very debatable pattern in real time frames.

So years of research is still an optimistic hope, especially since asymmetry rules, e.g. the same government debts can sometimes grow for decades:)

 
Mihail Marchukajtes:

Backing up the conversation, bros....

In the early days of becoming and acquiring popularity among the masses there was one of the fundamental rules comparable to the rule of input rubbish output and it sounds something like this "If a task can be solved without the help of neural networks, it should be solved in this way", i.e. the abbreviated meaning of the phrase: when a task has no direct or explicit solution only in that case it is reasonable to use NS. That is, NS is a last resort when solving problems of current or future uncertainty in complex areas, with an implicit solution, etc. But if the problem can be solved so.... without NS, then it should be solved that way.... without NS. Then the result of the solution will always be stable, while NS implies some freedom in solving.... like I want to do this today, and tomorrow I will want to do this.... As an example.

Unfortunately, maybe that's the reason I am so dumb and don't know much about IO, during all my career I have read only 2-3 books at the very beginning of my way, but no matter how many times I returned to IO literature it was always boring, because it was often written with things I already knew and I couldn't get anything new from it. Therefore, I have an interesting task which I will devote a separate topic to... So... everybody can do it, but I can't????

Probably about the same as they said about the use of computers in the early days of the big computer, and Gates disobeyed and created a personal computer in his garage with BASIC, and who would have thought then that the software would be played by children.

I think there will be something similar with NS too, AI for kids and housewives is already on the doorstep...

PS
But the topic of your site about car service in the Forex Expert Advisor section is out of place. Ask the moderator to move it to the general discussion, otherwise they may think that MetaQuotes and its EAs are already aspiring to the car business:)

 
Mihail Marchukajtes:

Backing up the conversation, bros....

In the early days of becoming and acquiring popularity among the masses there was one of the fundamental rules comparable to the rule of input rubbish output and it sounds something like this "If a task can be solved without the help of neural networks, it should be solved", i.e. the abbreviated meaning of the phrase: when a task has no direct or explicit solution, only in that case it is reasonable to use NS. That is, NS is a last resort when solving problems of current or future uncertainty in complex areas, with an implicit solution, etc. But if the problem can be solved so.... without NS, then it should be solved that way.... without NS. Then the result of the solution will always be stable, while NS implies some freedom in solving.... like I want to do this today, and tomorrow I will want to do this.... As an example.

Unfortunately, maybe that's the reason I am so dumb and don't know much about IO, during all my career I have read only 2-3 books at the very beginning of my way, but no matter how many times I returned to IO literature it was always boring, because it was often written with things I already knew and I couldn't get anything new from it. Therefore, I have an interesting task which I will devote a separate topic to... So... everybody can, but I can't????

Misha, you just need to understand once and for what purpose the NS is needed.

The thing is that some forum users have quite an algorithmic mindset.

It looks like this: if this, etc.

With this approach, there are a lot of options for solving a problem, in which most of the answers will be correct.

In order not to bother too much in searching for the optimal correct answer, the NS was invented. Except that neuronics algorithms are specifically sharpened for econometrics......

 

@Aleksei Lazo

Your last template is displaying with errors because part of the order chart objects are outside the bars, i.e. opening at non-existent prices:

maybe the problem is in the GMT server of the broker, I use the same one that I used when generating the EA from the previous template.

The second question concerns the trading principle used in the template, I am not an expert in martech technology, but

but it seems to me that we are using root-filling and the averaging of aggregate positions.

If so, it's not really the right technology for machine learning, at least I'm trying to train algorithms on

and identify some kind of pattern...

 
Ivan Negreshniy:

@Aleksei Lazo

Your last template is displaying with errors because part of the order chart objects are outside the bars, i.e. opening at non-existent prices:

maybe the problem is in the GMT server of the broker, I use the same one that I used when generating the EA from the previous template.

The second question concerns the trading principle used in the template, I am not an expert in martech technology, but

but it seems to me that we are using root-filling and the averaging of aggregate positions.

If so, it's not really a suitable technology for machine learning, at least I'm trying to train algorithms on

Analyse price patterns and identify any patterns...

Afternoon, could this be related to slippage what's on the price charts?

 
Ivan Negreshniy:

I propose a tool for automating pattern preparation, which is the makeSignals Expert Advisor that plots trading signals in the form of arrows on the chart itself.

Once the signals have been applied a trader can evaluate them, correct them by moving, removing or adding new ones, and then save it all to the template file (menu - Charts/Template/Save Template...).

The Expert Advisor has the following settings:

  • Signal bars count - number of bars on which the signal is calculated
  • Buy signal pips - estimated number of points of profit for the buy signal
  • Sell signal pips - calculated number of points of profit for the sell signal
  • Start date time - beginning of a period in which signals are calculated and applied
  • End date time - end date of the period in which signals are calculated and applied
  • Arrow drawind type - type of the graphical object - arrows used for drawing signals
  • Indicator used type - type of indicator used as a filter of signals
  • Clear all on exit - delete all graphical objects when disconnecting the Expert Advisor

The advisor searches within a given interval and plots on the chart all the signals that match the calculated parameters (number of bars and number of pips) and can also filter them if you select the indicator used so far available only two - indicator ZigZag and crossover of slow and fast EMA.

Information about the signals is displayed in the comment line - they are interval, size in points and current number of BUY and SELL signals, respectively.


I installed the filemakeSignals.mq4 in the Expert Advisor folder, updated it several times and reloaded MT5 several times, but this EA does not show up in "Expert Advisors" ... What is wrong? Why is it not seen?


 
btc.mmd:

InstalledmakeSignals.mq4 file in Expert Advisor folder, updated it several times, reloaded MT5 several times, this Expert Advisor did not appear in "Expert Advisors"... What's wrong, why is it not visible, where is it hiding?

mql4 file in MT5 ???

 
Vitaly Muzichenko:

mql4 file in mt5 ???

oops........ sorry!!!! ))))))))))))))), I fucked up, I must not have had enough ))))), thanks for the tip, I just wasn't paying attention ...

1...456789101112
New comment