Machine Learning: Optimal RSI [YinYangAlgorithms]

// This source code is subject to the terms of the Mozilla Public License 2.0 at https://mozilla.org/MPL/2.0/
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@    ,@@@@@@@@@@@@@@@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@      @@@@@@@@@@@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@        @@@@@@@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@         @@@@@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@           @@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@        .@@@@@@@@@@@@@@@            @@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@          *@@@@@@@@@@@@@@             @@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@         @@@@@@@@@@@@@@@               @@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@     @@@@@@@@@@@@@@@@                 @@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@                  @@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@.                    @@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@                      @@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@.                         @
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@                             @
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@,                                       @
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@                                                @
// @@@@@@@@@@@@@@@@@@@@@@@@@@@                                                    @
// @@@@@@@@@@@@@@@@@@@@@@@@@                                                     @@
// @@@@@@@@@@@@@@@@@@@@@@@                                                       @@
// @@@@@@@@@@@@@@@@@@@@@@                                                       @@@
// @@@@@@@@@@@@@@@@@@@@@*                @@@@@                                 @@@@
// @@@@@@@@@@@@@@@@@@@@@               @@@@@@@@@                              @@@@@
// @@@@@@@@@@@@@@@@@@@@@              @@@@@@@@@@@                           @@@@@@@
// @@@@@@@@@@@@@@@@@@@@@               @@@@@@@@%                           @@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@                                                @@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@                                            @@@@@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@                                        %@@@@@@@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@                                   @@@@@@@@@@@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@                           @@@@@@@@@@@@@@@@@@@@@@@@
// @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@                @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
// © YinYangAlgorithms

//@version=5
indicator("Machine Learning: Optimal RSI [YinYangAlgorithms]", max_bars_back=500)
// ~~~~~~~~~~~~ INPUTS ~~~~~~~~~~~~ //
//Optimal RSI
showSignals = input.bool(true, "Show Crossing Signals", group="RSI Settings", inline="rsiToggles")
showTables = input.bool(true, "Show Tables", group="RSI Settings", inline="rsiToggles")
showNewSettings = input.bool(true, "Show New Settings", group="RSI Settings", inline="rsiToggles")
showBollingerBands = input.bool(true, "Show Bollinger Bands", group="RSI Settings", tooltip="Show RSI Bollinger Bands. These bands work like the TDI Indicator, except its length changes as it uses the current RSI Optimal Length.")
optimalType = input.string("All Crossings", "Optimal RSI Type", options=["All Crossings", "Extremity Crossings"], group="RSI Settings", tooltip="This is how we calculate our Optimal RSI. Do we use all RSI and RSI MA Crossings or just when it crosses within the Extremities.")
aiAdjust = input.string("Auto", "Adjust Optimal RSI Lookback and RSI Count", options=["Auto", "Manual"], group="RSI Settings", tooltip="Auto means the script will automatically adjust the Optimal RSI Lookback Length and RSI Count based on the current Time Frame and Bar Index's on chart. This will attempt to stop the script from 'Taking too long to Execute'.\nManual means you have full control of the Optimal RSI Lookback Length and RSI Count.")
optimalLength = input.int(200, "Optimal RSI Lookback Length", maxval=500, minval=10, group="RSI Settings", tooltip="How far back are we looking to see which RSI length is optimal? Please note the more bars the lower this needs to be. For instance with BTC/USDT you can use 500 here on 1D but only 200 for 15 Minutes; otherwise it will timeout.")
rsiCount = input.int(30, "RSI Count", group="RSI Settings", maxval=50, minval=5, tooltip="How many lengths are we checking? For instance, if our 'RSI Minimum Length' is 4 and this is 30, the valid RSI lengths we check is 4-34.")
rsiMinLength = input.int(4, "RSI Minimum Length", group="RSI Settings", minval=1, tooltip="What is the RSI length we start our scans at? We are capped with RSI Count otherwise it will cause the Indicator to timeout, so we don't want to waste any processing power on irrelevant lengths.")
maLength = input.int(14, "RSI MA Length", group="RSI Settings", minval=1, tooltip="What length are we using to calculate the optimal RSI cross' and likewise plot our RSI MA with?")
backupLength = input.int(14, "Extremity Crossings RSI Backup Length", minval=1, group="RSI Settings", tooltip="When there is no Optimal RSI (if using Extremity Crossings), which RSI should we use instead?")
//Machine Learning
useRationalQuadratics = input.bool(true, "Use Rational Quadratics", group="Machine Learning", tooltip="Rationalizing our Close may be beneficial for usage within ML calculations.")
onlyUseSimilarMA = input.bool(false, "Filter RSI and RSI MA", group="Machine Learning", tooltip="Should we filter the RSI's before usage in ML calculations? Essentially should we only use RSI data that are of the same type as our Optimal RSI? For instance if our Optimal RSI is Bullish (RSI > RSI MA), should we only use ML RSI's that are likewise bullish?")
useMachineLearning = input.string("Simple Average", "Machine Learning Type", options=["KNN Average", "KNN Exponential Average", "Simple Average", "None"], group="Machine Learning", tooltip="Are we using a Simple ML Average, KNN Mean Average, KNN Exponential Average or None?")
distanceType = input.string("Both", "KNN Distance Type", options=["Both", "Max", "Min"], group="Machine Learning", tooltip="We need to check if distance is within the KNN Min/Max distance, which distance checks are we using.")
mlLength = input.int(10, "Machine Learning Length", minval=1, group="Machine Learning", tooltip="How far back is our Machine Learning going to keep data for.")
knnLength = input.int(3, "k-Nearest Neighbour (KNN) Length", minval=1, group="Machine Learning", tooltip="How many k-Nearest Neighbours will we account for?")
fastLength = input.int(1, "Fast ML Data Length", minval=1, group="Machine Learning", tooltip="What is our Fast ML Length? This is used with our Slow Length to create our KNN Distance.")
slowLength = input.int(5, "Slow ML Data Length", minval=2, group="Machine Learning", tooltip="What is our Slow ML Length? This is used with our Fast Length to create our KNN Distance.")

// ~~~~~~~~~~~~ VARIABLES ~~~~~~~~~~~~ //
distanceTypeMin = distanceType == "Min" or distanceType == "Both"
distanceTypeMax = distanceType == "Max" or distanceType == "Both"
adjustRatio = aiAdjust == "Auto"

// ~~~~~~~~~~~~ FUNCTIONS ~~~~~~~~~~~~ //
//Used to refactor INPUT Settings: Optimal RSI Lookback Length and RSI Count
    //When there are bar indexs due to timeframe than having these values too high will cause the indicator to timeout
autoAdjust() =>
    //Default settings (this should work on almost all Time Frames and Pairs (Premium or less Plans, not sure about > Premium))
    newLength = 90
    newCount = 20
    //total lookback can't be >= 50,000,000
        //how this is calculated is length * count * bar_index
        //IE. newLength (90) * newCount (25) * 24,000 (1 HR BTC/USDT Binance Index's available) = 54,000,000 ... this will timeout
        //The 50,000,000 lookback might also be based on Premium TradingView (40 second chart processing), so if you have a lower plan you may
            //need to adjust this accordingly
    //adjust based on bar_index, timeframe was an idea, but depending on what you're trading, the timeframe can still have too many indexs'
    if barstate.isrealtime
        //For realtime we no longer need to factor the Time Frame as we know how many bar_index's there are
        if bar_index <= 20000
            if bar_index <= 5000
                newLength := 200
                newCount := 30
            else if bar_index <= 10000
                newLength := 150
                newCount := 30
            else if bar_index <= 20000
                newLength := 125
                newCount := 25
            else
                newLength := 100
                newCount := 20
    else
        //For historical bar, we factor in the Time Frame as we don't know how many bar index's will be on this chart yet
        if bar_index <= 20000
            _timeS = timeframe.period
            _time = str.tonumber(_timeS)
            if _time >= 240 and _time < 720
                newLength := 125
                newCount := 25
            else if bar_index <= 10000 and _time >= 720 and _time < 1440
                newLength := 150
                newCount := 30
            else if bar_index <= 5000 and (str.contains(_timeS, "D") or str.contains(_timeS, "W") or str.contains(_timeS, "M"))
                newLength := 200
                newCount := 30
    [newLength, newCount]

//@jdehorty Kernel Function
//used to turn a source into a rational quadratic which performs better in ML calculations
rationalQuadratic(series float _src, simple int _lookback, simple float _relativeWeight, simple int startAtBar) =>
    float _currentWeight = 0.
    float _cumulativeWeight = 0.
    _size = array.size(array.from(_src))
    for i = 0 to _size + startAtBar
        y = _src[i]
        w = math.pow(1 + (math.pow(i, 2) / ((math.pow(_lookback, 2) * 2 * _relativeWeight))), -_relativeWeight)
        _currentWeight += y*w
        _cumulativeWeight += w
    yhat = _currentWeight / _cumulativeWeight
    yhat

//Consistent and easy way to fill a table cell
f_fillCell(_table, _column, _row, _value, _color) =>
    table.cell(_table, _column, _row, _value, bgcolor = color.new(_color, 75), text_color = _color, width = 20)

//Same as ta.sma but this way we can use a series length
pine_sma(x, series int y) =>
    sum = 0.0
    for i = 0 to y - 1
        sum := sum + x[i] / y
    sum

//Same as ta.rma but this way we can use a series length
pine_rma(src, series int length) =>
    alpha = 1/length
    sum = 0.0
    sum := na(sum[1]) ? pine_sma(src, length) : alpha * src + (1 - alpha) * nz(sum[1])

//Same as ta.rsi but this way we can use a series length
pine_rsi(x, series int y) =>
    u = math.max(x - x[1], 0) // upward ta.change
    d = math.max(x[1] - x, 0) // downward ta.change
    rs = pine_rma(u, y) / pine_rma(d, y)
    res = 100 - 100 / (1 + rs)
    res

//Calculate the optimal RSI length based on settings
getOptimalRSILength() =>
    //Storage to know the RSI and RSI MA cross percents (how much profit or loss has occured since this cross)
    crossPercents = array.new_float(rsiCount) //we use close not the rsi level
    //Get our source
    _src = useRationalQuadratics ? rationalQuadratic(close, 8, 8., 25) : close
    //Calculate our Optimal RSI
    for i = 0 to rsiCount - 1
        //scan all bar indexs for each RSI type to evaluate its crosses and percent increases
        len = i + rsiMinLength
        crossPercent = 0.
        crossType = 0 // 0 = none, -1 = under, 1 = over
        crossClose = 0. //close // was 0.
        crossCount = 0
        inExtremity = false
        for a = 0 to optimalLength - 1
            //get current bars RSI and MA
            currentRSI = pine_rsi(close[a], len)
            currentMA = pine_sma(currentRSI, maLength)
            //check for cross'
            crossOver = ta.crossover(currentRSI, currentMA)
            currentOver = optimalType == "All Crossings" or inExtremity ? crossOver :
             currentRSI <= 40 and crossOver
            crossUnder = ta.crossunder(currentRSI, currentMA)
            currentUnder = optimalType == "All Crossings" or inExtremity ? crossUnder :
             currentRSI >= 60 and crossUnder
            //calculate cross percent and update data
            if currentOver
                if crossType != 0
                    crossPercent += crossClose / close[a]
                    crossCount += 1
                crossClose := close[a]
                crossType := 1
                inExtremity := not inExtremity
            else if currentUnder
                if crossType != 0
                    crossPercent += close[a] / crossClose
                    crossCount += 1
                crossClose := close[a]
                crossType := -1
                inExtremity := not inExtremity
        //save the profit % of this RSI length
        crossPercents.set(i, crossPercent / crossCount)

    //See which RSI Length produced the highest profit
    bestPercent = -100000.
    bestIndex = 0
    for p = 0 to rsiCount - 1
        if crossPercents.get(p) > bestPercent
            bestPercent := crossPercents.get(p)
            bestIndex := p
    //get the optimal Length
    optimal = bestPercent != -100000 ? bestIndex + rsiMinLength : backupLength
    //return the best RSI Length
    [optimal, bestPercent]

//Get the exponential average of an array, where the exponential weight is focused on the first value of this array (current source)
getExponentialDataAverage(_data, _length) =>
    avg = 0.
    maxLen = math.min(_data.size(), _length)
    if maxLen > 0
        for i = 0 to maxLen - 1
            curData = _data.get(i)
            tempAvg = curData
            if i > 0
                for a = 0 to i
                    tempAvg += array.get(_data, a)
                tempAvg := math.avg(_data.get(0), tempAvg / i)
            avg += math.avg(_data.get(0), math.avg(curData, tempAvg))
        avg / _length
    else
        avg

//Uses KNN to sort distances with our ML fast and slow data
//This is a modified version that sorts distances but rather than saving and outputting the distance average it outputs the mean of the valid distance and averages the total
knnAverage_fromDistance(_dataFast, _dataSlow, _minDist, _maxDist) =>
    //failsafe we need at least 1 distance
    maxDist = not _minDist ? true : _maxDist
    //Calculate the distance between slow and fast moving ML Data
    distances = array.new_float(0)
    for i = 0 to _dataSlow.size() - 1
        distance = _dataSlow.get(i) - _dataFast.get(i)
        distances.push(distance)
    //clone the array so it doesn't modify it
    clonedDistances = distances.copy()
    //slice the length from the array and calculate the max value from this slice
    splicedDistances = clonedDistances.slice(0, math.min(knnLength, clonedDistances.size()))
    maxDistanceAllowed = splicedDistances.max()
    minDistanceAllowed = splicedDistances.min()
    //scan all distances and add any that are less than max distance
    validDistances = array.new_float(0)
    for i = 0 to distances.size() - 1
        if (not maxDist or distances.get(i) <= maxDistanceAllowed) and (not _minDist or distances.get(i) >= minDistanceAllowed)
            distAvg = (_dataSlow.get(i) + _dataFast.get(i)) / 2
            validDistances.push(distAvg)
    // Get exponential or regular average
    if useMachineLearning == "KNN Exponential Average"
        getExponentialDataAverage(validDistances, 1)
    else
        validDistances.avg()

// ~~~~~~~~~~~~ CALCULATIONS ~~~~~~~~~~~~ //
//Adjust length and counts
if adjustRatio
    [newLength, newCount] = autoAdjust()
    optimalLength := newLength
    rsiCount := newCount
//Get our Optimal RSI Length
[rsiLength, bestPercent] = getOptimalRSILength()
//Calculate our Optimal RSI
float optimalRSI = pine_rsi(close, rsiLength)
float rsi = optimalRSI
//Calculate our temp RSI MA (this will change if our RSI does due to ML Calculations)
tempMA = ta.sma(rsi, maLength)
//Calculate if the RSI is bullish (RSI >= RSI MA) or bearish (RSI < RSI MA)
rsiBull = rsi >= tempMA

//Apply Machine Learning logic to our Optimal RSI if selected
if useMachineLearning != "None"
    if useMachineLearning == "Simple Average"
        // -- A simple machine Learning Average
            //essentially just a simple Average of Optimal RSI data (potentially filtered to account for if its the same type (bullish or bearish))
        rsiData = array.new_float(0)
        for i = 0 to mlLength - 1
            simpleTempMa = pine_sma(optimalRSI[i], maLength)
            simpleTempBull = optimalRSI[i] > simpleTempMa
            if not onlyUseSimilarMA or simpleTempBull == rsiBull
                rsiData.push(optimalRSI[i])
        rsi := rsiData.avg()
    else
        // -- A still simple but more complex approach to Machine Learning using KNN to sort validDistances
        //Calculate our fast and slow MA's based on the current Optimal RSI
        float rsiFast = ta.sma(optimalRSI, fastLength)
        float rsiSlow = ta.sma(optimalRSI, slowLength)
        //create storage data for ML lookbacks
        rsiFastData = array.new_float(mlLength)
        rsiSlowData = array.new_float(mlLength)
        //populate our ML storage with lookbacks at our Fast and Slow Optimal RSIs
        for i = 0 to mlLength - 1
            rsiFastData.set(i, rsiFast[i])
            rsiSlowData.set(i, rsiSlow[i])
        //calculate our new Optimal RSI using KNN by using distances within KNN min/max and filtering by direction
        rsi := knnAverage_fromDistance(rsiFastData, rsiSlowData, distanceTypeMin, distanceTypeMax)

//Calculate our RSI MA (we do this later incase the RSI changed through Machine Learning Calculations)
rsiMA = ta.sma(rsi, maLength)
//Calculate RSI and RSI MA crosses (Signals)
bullCross = ta.crossover(rsi, rsiMA)
bearCross = ta.crossunder(rsi, rsiMA)
//Bollinger Bands
upper_inner = 0.
lower_inner = 0.
upper_outer = 0.
lower_outer = 0.
if showBollingerBands
    basis = ta.sma(rsi, optimalLength)
    dev_inner = 1.6185 * ta.stdev(rsi, optimalLength)
    dev_outer = 2.0 * ta.stdev(rsi, optimalLength)
    upper_inner := basis + dev_inner
    lower_inner := basis - dev_inner
    upper_outer := basis + dev_outer
    lower_outer := basis - dev_outer

// ~~~~~~~~~~~~ PLOTS ~~~~~~~~~~~~ //
//Bollinger Bands
bb_up_outer = plot(showBollingerBands ? upper_outer : na, "Upper Band", color = color.new(color.green, 40), linewidth = 1)
bb_up_inner = plot(showBollingerBands ? upper_inner : na, "Upper Band", color = color.new(color.green, 60), linewidth = 1)
fill(bb_up_outer, bb_up_inner, color=color.new(color.green, 95))
bb_down_outer = plot(showBollingerBands ? lower_outer : na, "Lower Band", color = color.new(color.red, 40), linewidth = 1)
bb_down_inner = plot(showBollingerBands ? lower_inner : na, "Lower Band", color = color.new(color.red, 60), linewidth = 1)
fill(bb_down_outer, bb_down_inner, color=color.new(color.red, 95))
//RSI Bands
h0 = hline(70, "Upper Band", color=#787B86)
hline(50, "Middle Band", color=color.new(#787B86, 50))
h1 = hline(30, "Lower Band", color=#787B86)
fill(h0, h1, color=color.rgb(126, 87, 194, 90), title="Background")
//RSI + RSI MA and gradient fills
rsiPlot = plot(rsi, "RSI", color=#7E57C2)
plot(rsiMA, "RSI MA", color=color.yellow)
midLinePlot = plot(50, color = na, editable = false, display = display.none)
fill(rsiPlot, midLinePlot, 100, 70, top_color = color.new(color.green, 0), bottom_color = color.new(color.green, 100), title = "RSI Overbought Gradient Fill")
fill(rsiPlot, midLinePlot, 30, 0, top_color = color.new(color.red, 100), bottom_color = color.new(color.red, 0), title = "RSI Oversold Gradient Fill")
//Crossing Signals
plot(showSignals and bullCross ? rsi : na, color=color.green, linewidth=3, style=plot.style_circles, title="RSI Bull Signal")
plot(showSignals and bearCross ? rsi : na, color=color.red, linewidth=3, style=plot.style_circles, title="RSI Bear Signal")

// ~~~~~~~~~~~~ TABLES ~~~~~~~~~~~~ //
var table perfTable = table.new(position.top_right, 4, 1, border_width = 5)
if showTables and barstate.islast
    f_fillCell(perfTable, 0, 0, "Optimal RSI Length: " + str.tostring(rsiLength), color.green)
    f_fillCell(perfTable, 1, 0, "Optimal Profit %: " + (bestPercent != -100000 ? str.tostring(math.round(bestPercent, 8)) : "None"), color.green)
    if showNewSettings and adjustRatio
        f_fillCell(perfTable, 2, 0, "New Lookback Length: " + str.tostring(optimalLength), color.green)
        f_fillCell(perfTable, 3, 0, "New RSI Count: " + str.tostring(rsiCount), color.green)

// ~~~~~~~~~~~~ ALERTS ~~~~~~~~~~~~ //
alertcondition(bullCross, title="RSI Bull Signal", message="RSI Bull Signal")
alertcondition(bearCross, title="RSI Bear Signal", message="RSI Bear Signal")

// ~~~~~~~~~~~~ END ~~~~~~~~~~~~ //