Neural networks programming with prorealtime

Leo · Fri, 17 Aug 2018 23:07:46 +0000

Hi all, I start to be interesting in Neural Network, it seems to be possible to do in prorealtime even though it has not Arrays (brace yourself, very long codes are coming). Now with multi-time-frame support and neural network, looks very interesting for trading. I found a very useful web that explain neural network in very simple way: http://neuralnetworksanddeeplearning.com/chap1.html I will start experimenting with this topic. For example, I code a Classifier for long entries. //Variables: //candlesback=5 //ProfitRiskRatio=2 //spread=1.5 myATR=average[20](range)//+std[20](range) ExtraStopLoss=MyATR //ExtraStopLoss=3*spread*pipsize classifier=0 //for long trades FOR scan=1 to candlesback DO IF classifier[scan]=1 then BREAK ENDIF LongTradeLength=ProfitRiskRatio*(close[scan]-(low[scan]-ExtraStopLoss[scan])) IF close[scan]+LongTradeLength < high-spread*pipsize then IF lowest[scan+1](low) > low[scan]-ExtraStopLoss[scan]+spread*pipsize then classifier=1 candleentry=barindex-scan BREAK ENDIF ENDIF NEXT IF classifier=1 then DRAWSEGMENT(candleentry,close[barindex-candleentry],barindex,close[barindex-candleentry]+LongTradeLength) COLOURED(0,250,0) DRAWELLIPSE(candleentry-1,low[barindex-candleentry]-ExtraStopLoss,barindex+1,high+ExtraStopLoss) COLOURED(0,250,0) ENDIF return

This topic has 126 replies, 8 voices, and was last updated 1 year ago by MobiusGrey.

Tagged: data mining, machine learning

Viewing 15 posts - 91 through 105 (of 127 total)

← 1 2 3 4 5 6 7 8 9 →

12/12/2018 at 8:46 AM #86915

Leo

Really odd. As I said, it can be similar when the predictions is very good. But exactly the same. I do not know why. Please attached your file with your update. Today night or tomorrow morning I will have a look at it.

Please try loading more bars in the chart or try another time frame or try another instrument. Are my network and your network having the same output?

Leo

Participant

Topics: 17

Replies: 357

Been thanked: 144 times

12/12/2018 at 10:06 PM #86973

Leo

Hi Stefu102,

I check your neural network and it look everything normal.

I loaded 15min time frame dax, 20000 bars

This are the variables I corrected according to previous posts

Trend001= (Average[50](close) + (2.5 * STD[50](close)) -close) / close
Trend002= (close - Average[50](close) - (2.5 * STD[50](close)) ) /close
Trend003= ( close- (highest[9](close)+lowest[9](close))/2 ) / close
Trend004= (close - (highest[26](close)+lowest[26](close))/2 ) / close

Trend001= (Average[50](close) + (2.5 * STD[50](close)) -close) / close

Trend002= (close - Average[50](close) - (2.5 * STD[50](close)) ) /close

Trend003= ( close- (highest[9](close)+lowest[9](close))/2 ) / close

Trend004= (close - (highest[26](close)+lowest[26](close))/2 ) / close

See also the photo I attached.

You said that you where testing with 25 ticks bars… To be honest I have never worked with ticks bars and I do not know the behaviour of the code in those graphs.

Cheers

Leo

Participant

Topics: 17

Replies: 357

Been thanked: 144 times

12/12/2018 at 10:31 PM #86975

stefou102

Hello Leo,

ok I see now that I didn’t pay attention to the value of the different lines of the visual indicator, only to the red and green bar giving the signal. What remains weird to me is that the green and red bar are the same, even if the input is different. I should have expected more variation in order to gain confidence in the neural network as a whole

stefou102

Participant

Topics: 22

Replies: 171

Been thanked: 71 times

12/12/2018 at 10:40 PM #86976

Leo

Well… For me look normal. The Neural Network learns and make the prediction. My indicators and your indicators provide signals and the adgoritm has found a correlation.

Both Neural networks predict that it will continues the upward trend… And if you see the chart it look pretty upward trend.

Look awesome for me.

Leo

Participant

Topics: 17

Replies: 357

Been thanked: 144 times

12/12/2018 at 11:10 PM #86977

stefou102

Sorry to keep asking for your explanation, and thx for your reply. As you can see in the image attached, the value of the lines “prediction short/long” are pretty different from one network to another, but as said the red/green signal of the classifier are always the same, for the whole historical dataset, be it 10,000 or 100,000. So I remain skeptic about the classifier…In the image attached, I have the lines prediction long and short that have a value of 0, but still have the classifier trigger to 1…

COuld you also explain what is your parameter “betha”?

stefou102

Participant

Topics: 22

Replies: 171

Been thanked: 71 times

12/13/2018 at 6:03 AM #86984

Leo

Now I understand your issue, the thing is:

The classifier is totally independent of the network

As long as the classifier has the same parameters their output is the same.

The classifier provides the data for the neural network to learn.

The bar appears when the classifier detect a fast movement of 50 pips.

About your other question: what is the betha?

Search in the Web “gradient descent with momentum”. It is a method for increase the learning process speed. Betha is like a correction when the convergence direction changes and accelerate when the direction is the correct one. There are other methods but I find very complex to code in ProRealTime.

By the way. About your variables. Multiply them for 100 then you have a bigger number for avoid things related to presicion.

Hope I clarify your questions. Keep in mind that I am newbie in this Artifitial Intelligent topic.

Leo

Participant

Topics: 17

Replies: 357

Been thanked: 144 times

12/13/2018 at 7:29 PM #87054

stefou102

Ok now that I have multiply the variables by 100, I have different values for the classifier as well!

I’m wondering what kind of variable can we play with. Do you think bolean value would work? like simply if the previous bar was positive or negative

stefou102

Participant

Topics: 22

Replies: 171

Been thanked: 71 times

12/13/2018 at 7:37 PM #87055

stefou102

I mean for the bolean, we need to create a separate indicator and call it in the neural network, right?

stefou102

Participant

Topics: 22

Replies: 171

Been thanked: 71 times

12/13/2018 at 8:21 PM #87058

stefou102

Sorry please disregard my comment about the classifier. I misunderstood because I thought that the red/green histogram bars were buy/sell signals, while they are only the result of the classifier having spotted a past winning move on which the neural network can learn.

stefou102

Participant

Topics: 22

Replies: 171

Been thanked: 71 times

12/14/2018 at 5:53 AM #87070

Leo

HI,

I play with the following variables…

First add a new parameter for choosing or simply copy paste one set

IF SetParameters=1 THEN  
// STOCHASTIC
PseudoStochastic1= (close-lowest[2](low[0])) / (highest[2](high[0])-lowest[2](low[0]))
PseudoStochastic2= (close-lowest[3](low[0])) / (highest[3](high[0])-lowest[3](low[0]))
PseudoStochastic3= (close-lowest[5](low[0])) / (highest[5](high[0])-lowest[4](low[0]))
PseudoStochastic4= (close-lowest[8](low[0])) / (highest[8](high[0])-lowest[8](low[0]))

variable1= PseudoStochastic1
variable2= PseudoStochastic2
variable3= PseudoStochastic3
variable4= PseudoStochastic4
ELSIF SetParameters=2 THEN 
//MOVING AVERAGES
ratio1= 100*(close-average[7](close))/average[7](close)
ratio2= 100*(close-average[13](close))/average[13](close)
ratio3= 100*(close-average[34](close))/average[34](close)
ratio4= 100*(close-average[77](close))/average[77](close)

variable1= ratio1
variable2= ratio2
variable3= ratio3
variable4= ratio4
ELSIF SetParameters=3 THEN
//STANDARD INDICATORS
indicator1=RSI[2](close)
indicator2=RSI[14](close)
indicator3=MACD[12,26,9](close)
indicator4=ADX[14]

variable1= indicator1
variable2= indicator2
variable3= indicator3
variable4= indicator4
else
//BOLLINGER BANDS
SMA001=average[20](close)
STD001=STD[20](close)
SMA002=average[200](close)
STD002=STD[200](close)
ANGLESMA001= ATAN( (SMA001-SMA001[1])  /pipsize ) * 180 / 3.1416
ANGLESMA002= ATAN( (SMA002-SMA002[1])  /pipsize ) * 180 / 3.1416
variable1=(close-SMA001)/STD001
variable2=(close-SMA002)/STD002
variable3=ANGLESMA001
variable4=ANGLESMA002
ENDIF

IF SetParameters=1 THEN

// STOCHASTIC

PseudoStochastic1= (close-lowest[2](low[0])) / (highest[2](high[0])-lowest[2](low[0]))

PseudoStochastic2= (close-lowest[3](low[0])) / (highest[3](high[0])-lowest[3](low[0]))

PseudoStochastic3= (close-lowest[5](low[0])) / (highest[5](high[0])-lowest[4](low[0]))

PseudoStochastic4= (close-lowest[8](low[0])) / (highest[8](high[0])-lowest[8](low[0]))

variable1= PseudoStochastic1

variable2= PseudoStochastic2

variable3= PseudoStochastic3

variable4= PseudoStochastic4

ELSIF SetParameters=2 THEN

//MOVING AVERAGES

ratio1= 100*(close-average[7](close))/average[7](close)

ratio2= 100*(close-average[13](close))/average[13](close)

ratio3= 100*(close-average[34](close))/average[34](close)

ratio4= 100*(close-average[77](close))/average[77](close)

variable1= ratio1

variable2= ratio2

variable3= ratio3

variable4= ratio4

ELSIF SetParameters=3 THEN

//STANDARD INDICATORS

indicator1=RSI[2](close)

indicator2=RSI[14](close)

indicator3=MACD[12,26,9](close)

indicator4=ADX[14]

variable1= indicator1

variable2= indicator2

variable3= indicator3

variable4= indicator4

else

//BOLLINGER BANDS

SMA001=average[20](close)

STD001=STD[20](close)

SMA002=average[200](close)

STD002=STD[200](close)

ANGLESMA001= ATAN( (SMA001-SMA001[1]) /pipsize ) * 180 / 3.1416

ANGLESMA002= ATAN( (SMA002-SMA002[1]) /pipsize ) * 180 / 3.1416

variable1=(close-SMA001)/STD001

variable2=(close-SMA002)/STD002

variable3=ANGLESMA001

variable4=ANGLESMA002

ENDIF

HOPE YOU LIKE THEM 🙂

1 user thanked author for this post.

Raspoutine

Leo

Participant

Topics: 17

Replies: 357

Been thanked: 144 times

12/14/2018 at 4:26 PM #87112

TACBOLSA

Siento el Idioma pero para poder entendernos mejor os lo paso en Español..

Llevo trabajando uno tiempo en un Indicador red Neuronal, basado en un indicador de Timing que tengo… lo he combinado con vuestras aportaciones y creo que da unos resultados a tener en cuenta.

Lo probare y tendré en pruebas para ver como actúa.. si obtenemos datos interesantes combinamos los codigos y lo publicamos..

TACBOLSA

Participant

Topics: 7

Replies: 36

Been thanked: 35 times

12/14/2018 at 4:38 PM #87114

Vonasi

English only in the English speaking forums please.

If you want to communicate in Spanish on a topic then create a post in the Spanish forums with a link to the English topic.

Inglés solo en los foros de habla inglesa por favor.

Si desea comunicarse en español sobre un tema, cree una publicación en los foros en español con un enlace al tema en inglés.

Vonasi

Moderator

Topics: 192

Replies: 5088

Been thanked: 1811 times

12/14/2018 at 5:06 PM #87119

Leo

Hi TAC bolsa…

Sounds great… if think the lenguage is an issue then open a new Forum and we discuss it there. Spanish is my mother tongue

Leo

Participant

Topics: 17

Replies: 357

Been thanked: 144 times

12/14/2018 at 5:32 PM #87125

TACBOLSA

Language is not a problem.

If I get good results, I mention it in the Spanish forum.

TACBOLSA

Participant

Topics: 7

Replies: 36

Been thanked: 35 times

12/14/2018 at 5:47 PM #87129

Leo

Can you please copy the link to the Forum, thanks.

Leo

Participant

Topics: 17

Replies: 357

Been thanked: 144 times

Author

Posts

Viewing 15 posts - 91 through 105 (of 127 total)

← 1 2 3 4 5 6 7 8 9 →

Find exclusive trading pro-tools on