Conversion of indicator Volume SuperTrend AI from the TradingView platform

Ξ3 · Mon, 26 Feb 2024 07:10:46 +0000

Hi Nicholas, Would it be possible to convert the Volume SuperTrend AI indicator to ProRealTime code? It would be greatly appreciated. The details of this indicator can be found at: https://www.tradingview.com/script/eTgP2ymK-Volume-SuperTrend-AI-Expo/ █ Overview The Volume SuperTrend AI is an advanced technical indicator used to predict trends in price movements by utilizing a combination of traditional SuperTrend calculation and AI techniques, particularly the k-nearest neighbors (KNN) algorithm. The Volume SuperTrend AI is designed to provide traders with insights into potential market trends, using both volume-weighted moving averages (VWMA) and the k-nearest neighbors (KNN) algorithm. By combining these approaches, the indicator aims to offer more precise predictions of price trends, offering bullish and bearish signals. Many thanks. ——-TRADINGVIEW CODE BELOW—————————————————————————————– // This work is licensed under a Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) https://creativecommons.org/licenses/by-nc-sa/4.0/ // © Zeiierman //@version=5 indicator(“Volume SuperTrend AI (Expo)”, overlay=true) // ~~ ToolTips { t1=”Number of nearest neighbors in KNN algorithm (k): Increase to consider more neighbors, providing a more balanced view but possibly smoothing out local patterns. Decrease for fewer neighbors to make the algorithm more responsive to recent changes. \n\nNumber of data points to consider (n): Increase for more historical data, providing a broader context but possibly diluting recent trends. Decrease for less historical data to focus more on recent behavior.” t2=”Length of weighted moving average for price (KNN_PriceLen): Higher values create a smoother price line, influencing the KNN algorithm to be more stable but less sensitive to short-term price movements. Lower values enhance responsiveness in KNN predictions to recent price changes but may lead to more noise. \n\nLength of weighted moving average for SuperTrend (KNN_STLen): Higher values lead to a smoother SuperTrend line, affecting the KNN algorithm to emphasize long-term trends. Lower values make KNN predictions more sensitive to recent SuperTrend changes but may result in more volatility.” t3=”Length of the SuperTrend (len): Increase for a smoother trend line, ideal for identifying long-term trends but possibly ignoring short-term fluctuations. Decrease for more responsiveness to recent changes but risk of more false signals. \n\nMultiplier for ATR in SuperTrend calculation (factor): Increase for wider bands, capturing larger price movements but possibly missing subtle changes. Decrease for narrower bands, more sensitive to small shifts but risk of more noise.” t4=”Type of moving average for SuperTrend calculation (maSrc): Choose based on desired characteristics. SMA is simple and clear, EMA emphasizes recent prices, WMA gives more weight to recent data, RMA is less sensitive to recent changes, and VWMA considers volume.” t5=”Color for bullish trend (upCol): Select to visually identify upward trends. \n\nColor for bearish trend (dnCol): Select to visually identify downward trends.\n\nColor for neutral trend (neCol): Select to visually identify neutral trends.” //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Input settings for K and N values k = input.int(3, title = “Neighbors”, minval=1, maxval=100,inline=”AI”, group=”AI Settings”) n_ = input.int(10, title =”Data”, minval=1, maxval=100,inline=”AI”, group=”AI Settings”, tooltip=t1) n = math.max(k,n_) //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Input settings for prediction values KNN_PriceLen = input.int(20, title=”Price Trend”, minval=2, maxval=500, step=10,inline=”AITrend”, group=”AI Trend”) KNN_STLen = input.int(100, title=”Prediction Trend”, minval=2, maxval=500, step=10, inline=”AITrend”, group=”AI Trend”, tooltip=t2) aisignals = input.bool(true,title=”AI Trend Signals”,inline=”signal”, group=”AI Trend”) Bullish_col = input.color(color.lime,””,inline=”signal”, group=”AI Trend”) Bearish_col = input.color(color.red,””,inline=”signal”, group=”AI Trend”) //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Define SuperTrend parameters len = input.int(10, “Length”, minval=1,inline=”SuperTrend”, group=”Super Trend Settings”) factor = input.float(3.0,step=.1,inline=”SuperTrend”, group=”Super Trend Settings”, tooltip=t3) maSrc = input.string(“WMA”,”Moving Average Source”,[“SMA”,”EMA”,”WMA”,”RMA”,”VWMA”],inline=””, group=”Super Trend Settings”, tooltip=t4) upCol = input.color(color.lime,”Bullish Color”,inline=”col”, group=”Super Trend Coloring”) dnCol = input.color(color.red,”Bearish Color”,inline=”col”, group=”Super Trend Coloring”) neCol = input.color(color.blue,”Neutral Color”,inline=”col”, group=”Super Trend Coloring”, tooltip=t5) //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Calculate the SuperTrend based on the user’s choice vwma = switch maSrc “SMA” => ta.sma(close*volume, len) / ta.sma(volume, len) “EMA” => ta.ema(close*volume, len) / ta.ema(volume, len) “WMA” => ta.wma(close*volume, len) / ta.wma(volume, len) “RMA” => ta.rma(close*volume, len) / ta.rma(volume, len) “VWMA” => ta.vwma(close*volume, len) / ta.vwma(volume, len) atr = ta.atr(len) upperBand = vwma + factor * atr lowerBand = vwma – factor * atr prevLowerBand = nz(lowerBand[1]) prevUpperBand = nz(upperBand[1]) lowerBand := lowerBand > prevLowerBand or close[1] < prevLowerBand ? lowerBand : prevLowerBand upperBand := upperBand < prevUpperBand or close[1] > prevUpperBand ? upperBand : prevUpperBand int direction = na float superTrend = na prevSuperTrend = superTrend[1] if na(atr[1]) direction := 1 else if prevSuperTrend == prevUpperBand direction := close > upperBand ? -1 : 1 else direction := close < lowerBand ? 1 : -1 superTrend := direction == -1 ? lowerBand : upperBand //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Collect data points and their corresponding labels price = ta.wma(close,KNN_PriceLen) sT = ta.wma(superTrend,KNN_STLen) data = array.new_float(n) labels = array.new_int(n) for i = 0 to n – 1 data.set(i, superTrend[i]) label_i = price[i] > sT[i] ? 1 : 0 labels.set(i, label_i) //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Define a function to compute distance between two data points distance(x1, x2) => math.abs(x1 – x2) //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Define the weighted k-nearest neighbors (KNN) function knn_weighted(data, labels, k, x) => n1 = data.size() distances = array.new_float(n1) indices = array.new_int(n1) // Compute distances from the current point to all other points for i = 0 to n1 – 1 x_i = data.get(i) dist = distance(x, x_i) distances.set(i, dist) indices.set(i, i) // Sort distances and corresponding indices in ascending order // Bubble sort method for i = 0 to n1 – 2 for j = 0 to n1 – i – 2 if distances.get(j) > distances.get(j + 1) tempDist = distances.get(j) distances.set(j, distances.get(j + 1)) distances.set(j + 1, tempDist) tempIndex = indices.get(j) indices.set(j, indices.get(j + 1)) indices.set(j + 1, tempIndex) // Compute weighted sum of labels of the k nearest neighbors weighted_sum = 0. total_weight = 0. for i = 0 to k – 1 index = indices.get(i) label_i = labels.get(index) weight_i = 1 / (distances.get(i) + 1e-6) weighted_sum += weight_i * label_i total_weight += weight_i weighted_sum / total_weight //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Classify the current data point current_superTrend = superTrend label_ = knn_weighted(data, labels, k, current_superTrend) //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Plot col = label_ == 1?upCol:label_ == 0?dnCol:neCol plot(current_superTrend, color=col, title=”Volume Super Trend AI”) upTrend = plot(superTrend==lowerBand?current_superTrend:na, title=”Up Volume Super Trend AI”, color=col, style=plot.style_linebr) Middle = plot((open + close) / 2, display=display.none, editable=false) downTrend = plot(superTrend==upperBand?current_superTrend:na, title=”Down Volume Super Trend AI”, color=col, style=plot.style_linebr) fill_col = color.new(col,90) fill(Middle, upTrend, fill_col, fillgaps=false,title=”Up Volume Super Trend AI”) fill(Middle, downTrend, fill_col, fillgaps=false, title=”Down Volume Super Trend AI”) //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Ai Super Trend Signals Start_TrendUp = col==upCol and (col[1]!=upCol or col[1]==neCol) and aisignals Start_TrendDn = col==dnCol and (col[1]!=dnCol or col[1]==neCol) and aisignals TrendUp = direction == -1 and direction[1] == 1 and label_ == 1 and aisignals TrendDn = direction == 1 and direction[1] ==-1 and label_ == 0 and aisignals plotshape(Start_TrendUp?superTrend:na, location=location.absolute, style= shape.circle, size=size.tiny, color=Bullish_col, title=”AI Bullish Trend Start”) plotshape(Start_TrendDn?superTrend:na, location=location.absolute, style= shape.circle,size=size.tiny, color=Bearish_col, title=”AI Bearish Trend Start”) plotshape(TrendUp?superTrend:na, location=location.absolute, style= shape.triangleup, size=size.small, color=Bullish_col, title=”AI Bullish Trend Signal”) plotshape(TrendDn?superTrend:na, location=location.absolute, style= shape.triangledown,size=size.small, color=Bearish_col, title=”AI Bearish Trend Signal”) //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~} // ~~ Alerts { alertcondition(Start_TrendUp, title =”1 Bullish Trend Start”, message = “AI Bullish Trend Start”) alertcondition(Start_TrendDn, title =”2 Bearish Trend Start”, message = “AI Bearish Trend Start”) alertcondition((Start_TrendUp or Start_TrendDn), title =”3 Any Trend Start”, message=”Any AI Trend Start”) alertcondition(TrendUp, title = “4 Bullish Trend Signal”, message = “AI Bullish Trend Signal”) alertcondition(TrendDn, title = “5 Bearish Trend Signal”, message = “AI Bearish Trend Signal”) alertcondition((TrendUp or TrendDn),title = “6 Any Trend Signal”, message =”Any AI Trend Signal”) //~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~}

This topic has 22 replies, 6 voices, and was last updated 1 month ago by Ξ3.

Viewing 8 posts - 16 through 23 (of 23 total)

← 1 2

03/07/2024 at 11:59 AM #229383

Acivi90

My first check was on DAX, useing IG.

I use 200 K bars for the exempel above.

thanks for your help!

Acivi90

Participant

Topics: 0

Replies: 7

Been thanked: 0 times

03/07/2024 at 12:52 PM #229389

PeterSt

I don’t know what is going on. So my test this morning was with PRT-IB. All fine (but poor result). Now with IG, I don’t get anything out of it; tried 1m, 5m and 5s timeframes. But wait … with 5s I do get trades. All on Feb 9 2024 which is the first day with 5s and 200K bars.

I noticed these things (maybe for Irván) : In PRT-IB it is too fast to be true (like 5 seconds), with the array manipulation. At least that is my feeling. In PRT-IG it takes extremely long (like a few minutes) which could be due to more ticks with IG (that IG bears more ticks is a fact). Now, seeing this result with quite some trades on that first day, it looks like something chokes. I thought of tooo many array elements used (max is 1 million AFAK), but that can’t be it after one day – I think.

I am afraid I have to hand this task to someone else because the time it takes to test it on IG is too long for me. Combined with the strange results (IB looking fine, apart from the very poor result) this is nothing to dive into, when it is not one’s own. Best advice could be to drop this all together and start over with another means of “SuperTrend”. There are some in the Library or just in the forum.
Sorry …

PeterSt

Participant

Topics: 70

Replies: 2287

Been thanked: 695 times

03/07/2024 at 1:24 PM #229393

PeterSt

With 5 seconds TF it always does one day only (just the first day where backtesting starts). Even if I change the last part to this :

If C1 then //AND timeEnterBefore AND timeEnterAfter AND not daysForbiddenEntry then
  Buy 1 contract at Market
endif
iF c2 THEN
  SELL AT MARKET
ENDIF

SET STOP %loss 0.3
SET TARGET %PROFIT 0.3

If C1 then //AND timeEnterBefore AND timeEnterAfter AND not daysForbiddenEntry then

Buy 1 contract at Market

endif

iF c2 THEN

SELL AT MARKET

ENDIF

SET STOP %loss 0.3

SET TARGET %PROFIT 0.3

this is so.

PeterSt

Participant

Topics: 70

Replies: 2287

Been thanked: 695 times

03/08/2024 at 2:37 PM #229475

JC_Bywan

Hi,

I had a look, found a few things but there is still something weird about upperband and lowerband calculations. Anyway, starting with a couple of typos first in the indicator script (which have therefore been carried forward into the proorder code too and also need to be changed there):

To be on same tune, line numbers below are as per code listing in post #229004 of previous page:

line 34: lowerdband to be replaced by: lowerband

line 50: directon (missing letter i) to be replaced by: direction

Second, “myindex” defined in line 101 didn’t seem to be used anywhere else so it caught my eye, I am not a tradingview expert, but it feels like probably one “i” at least would need to be replaced by this “myindex” variable to make the probuilder code behave the same as the tradingview code, I’d say at least in line 102:

line 102: labeli = $labels[i] to be replaced by: labeli = $labels[myindex] (to be confirmed by those familiar with tradingview)

Next, even with these changes, (and even when modifying, just in case, line 28 into: upperband = vwma + factor*atr

and line 34 into: lowerband = vwma – factor*atr ),

upperband and lowerband calculations seem to not work properly in the backtest (seen by graphing them and comparing them with indicator version where added to return line). So, comparing how it starts and seeing where from behaviours diverge is the next thing. This requires adding a line: defparam PRELOADBARS =0 in the backtest to compare exact same starts (otherwise proorder code starts earlier in history than probuilder code). It seems that behaviour is the same on intraday bars of first day only, but at the first intraday bar with volume of the second day, that is when backtest version doesn’t update upperband/lowerband at all, where the indicator version does… not seeing instantly why…

4 users thanked author for this post.

Ξ3,

WhyAskOZ,

Acivi90,

PeterSt

JC_Bywan

Moderator

Topics: 20

Replies: 2549

Been thanked: 839 times

03/08/2024 at 8:03 PM #229494

JC_Bywan

Ok,

For vwma we have a division with an average of volume in denominator. Where PRT trading with Ibkr has real exchange data with volume at each candle, IG is OTC and has ovn candles with no volume, so you see it coming before I finish the sentence…: no issue calculating vwma with Ibkr, but on IG for the ovn candles before first non-zero volume morning candle for small intraday timeframe, average of 0 volume on a period shorter than ovn length is zero, and vwma having a division by zero is undefined, cascading through issues that wouldn’t appear with Ibkr. That’s one part of the mistery solved.

From there, why the further differences in upperband and lowerband between probuilder code and proorder code within an IG environment? My best guess is that when an if statement has an undefined value in its condition (here, the “undefined” is upperband[1] at first intraday bar with non-zero volume in line 27, and lowerband[1] in line 33), such if statement would be handled differently between probuilder and proorder: the probuilder code would go into the “else” but keep the defined earlier values given in lines 18-19, the proorder code goes into the “else” and propagates for upperband=upperband[1] and lowerband=lowerband [1] a recurringly undefined value.

So I tried to workaround this by adding to the if statements for lines 27 and 33 in the proorder version this further condition to try to emulate the probuilder behaviour: (volume<>0 and volume[1]=0) (again taking as reference post #229004 of previous page for probuilder, actual matching lines in proorder version would have a different number)

Proorder equivalent to probuilder version Lines 27-28 become:

if upperband < upperband[1] or close[1] > upperband[1] or (volume<>0 and volume[1]=0) then
   upperband = vwma + factor*atr

1 2	if upperband < upperband[1] or close[1] > upperband[1] or (volume<>0 and volume[1]=0) then upperband = vwma + factor*atr

Proorder equivalent to probuilder version Lines 33-34 become:

if lowerband > lowerband[1] or close[1] < lowerband[1] or (volume<>0 and volume[1]=0) then
   lowerband = vwma – factor*atr

1 2	if lowerband > lowerband[1] or close[1] < lowerband[1] or (volume<>0 and volume[1]=0) then lowerband = vwma – factor*atr

It seems to give trades from first to last day, although I haven’t tried several assets and timeframes, and didn’t check if trades are correct, only that they are – at least and at last – there… I let those who want a proorder version to further check from there…

3 users thanked author for this post.

Ξ3,

Iván,

Acivi90

JC_Bywan

Moderator

Topics: 20

Replies: 2549

Been thanked: 839 times

03/11/2024 at 8:25 AM #229565

Iván

thank you very much for this magnificent contribution!

Iván

Moderator

Topics: 0

Replies: 306

Been thanked: 84 times

03/11/2024 at 9:23 AM #229579

WhyAskOZ

hi, Guys

I noticed since last I commented on this thread there are various expert made minor changes to various lines.

Would it be possible that someone (as I am not that expert in this field yet) compile all correct changes and add publish this code (Indicator and strategy separately) here and also on the library? That will be very useful for future use?

Thanks, all participant.

WhyAskOZ

Participant

Topics: 6

Replies: 32

Been thanked: 1 time

03/14/2024 at 7:15 AM #229744

Ξ3

Thank you to everyone for your contributions!

Ξ3

Participant

Topics: 12

Replies: 26

Been thanked: 3 times

Author

Posts

Viewing 8 posts - 16 through 23 (of 23 total)

← 1 2

Find exclusive trading pro-tools on