improve //@version=5 strategy("Machine Learning Adaptive SuperTrend [AlgoAlpha]",…
- User
improve
//@version=5 strategy("Machine Learning Adaptive SuperTrend [AlgoAlpha]", "AlgoAlpha - 🤖 Adaptive SuperTrend", overlay = true, max_labels_count = 500)
import TradingView/ta/7
atr_len = input.int(10, "ATR Length", group = "SuperTrend Settings") fact = input.float(3, "SuperTrend Factor", group = "SuperTrend Settings") training_data_period = input.int(100, "Training Data Length", group = "K-Means Settings") highvol = input.float(0.75, "Initial High volatility Percentile Guess", maxval = 1, group = "K-Means Settings", tooltip = "The initial guess of where the potential 'high volatility' area is, a value of 0.75 will take the 75th percentile of the range of ATR values over the training data period") midvol = input.float(0.5, "Initial Medium volatility Percentile Guess", maxval = 1, group = "K-Means Settings", tooltip = "The initial guess of where the potential 'medium volatility' area is, a value of 0.5 will take the 50th percentile of the range of ATR values over the training data period") lowvol = input.float(0.25, "Initial Low volatility Percentile Guess", maxval = 1, group = "K-Means Settings", tooltip = "The initial guess of where the potential 'low volatility' area is, a value of 0.25 will take the 25th percentile of the range of ATR values over the training data period") green = input.color(#00ffbb, "Bullish Color", group = "Appearance") red = input.color(#ff1100, "Bearish Color", group = "Appearance")
pine_supertrend(factor, atr) => src = hl2 upperBand = src + factor * atr lowerBand = src - 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 [superTrend, _direction]volatility = ta.atr(atr_len)
upper = ta.highest(volatility, training_data_period) lower = ta.lowest(volatility, training_data_period)
high_volatility = lower + (upper-lower) * highvol medium_volatility = lower + (upper-lower) * midvol low_volatility = lower + (upper-lower) * lowvol
iterations = 0
size_a = 0 size_b = 0 size_c = 0
hv = array.new_float() mv = array.new_float() lv = array.new_float() amean = array.new_float(1,high_volatility) bmean = array.new_float(1,medium_volatility) cmean = array.new_float(1,low_volatility)
if nz(volatility) > 0 and bar_index >= training_data_period-1
while ((amean.size() == 1 ? true : (amean.first() != amean.get(1))) or (bmean.size() == 1 ? true : (bmean.first() != bmean.get(1))) or (cmean.size() == 1 ? true : (cmean.first() != cmean.get(1)))): hv.clear() mv.clear() lv.clear() for i = training_data_period-1 to 0 if math.abs(volatility[i] - amean.first()) < math.abs(volatility[i] - bmean.first()) and math.abs(volatility[i] - amean.first()) < math.abs(volatility[i] - cmean.first()) hv.unshift(volatility[i]) if math.abs(volatility[i] - bmean.first()) < math.abs(volatility[i] - amean.first()) and math.abs(volatility[i] - bmean.first()) < math.abs(volatility[i] - cmean.first()) mv.unshift(volatility[i]) if math.abs(volatility[i] - cmean.first()) < math.abs(volatility[i] - amean.first()) and math.abs(volatility[i] - cmean.first()) < math.abs(volatility[i] - bmean.first()) lv.unshift(volatility[i]) amean.unshift(hv.avg()) bmean.unshift(mv.avg()) cmean.unshift(lv.avg()) size_a := hv.size() size_b := mv.size() size_c := lv.size() iterations := iterations + 1hv_new = amean.first() mv_new = bmean.first() lv_new = cmean.first() vdist_a = math.abs(volatility - hv_new) vdist_b = math.abs(volatility - mv_new) vdist_c = math.abs(volatility - lv_new)
distances = array.new_float() centroids = array.new_float()
distances.push(vdist_a) distances.push(vdist_b) distances.push(vdist_c)
centroids.push(hv_new) centroids.push(mv_new) centroids.push(lv_new)
cluster = distances.indexof(distances.min()) // 0 for high, 1 for medium, 2 for low assigned_centroid = cluster == -1 ? na : centroids.get(cluster)
[ST, dir] = pine_supertrend(fact, assigned_centroid)
// Define buy and sell conditions longCondition = ta.crossunder(dir, 0) shortCondition = ta.crossover(dir, 0)
// Execute buy and sell orders if (longCondition) strategy.entry("Long", strategy.long)
if (shortCondition) strategy.entry("Short", strategy.short)
// Stop Loss and Take Profit based on ATR atrVal = ta.atr(14) strategy.exit("Exit Long", from_entry = "Long", stop = close - 1.5 * atrVal, limit = close + 2 * atrVal) strategy.exit("Exit Short", from_entry = "Short", stop = close + 1.5 * atrVal, limit = close - 2 * atrVal)
t1 = 70 t2 = 95
upTrend = plot(close > ST ? ST : na, color = color.new(green, t1), style = plot.style_linebr) downTrend = plot(close < ST ? ST : na, color = color.new(red, t1), style = plot.style_linebr, force_overlay = false) bodyMiddle = plot(barstate.isfirst ? na : (open + close) / 2, "Body Middle", display = display.none)
fill(bodyMiddle, upTrend, (open + close) / 2, ST, color.new(green, t2), color.new(green, t1)) fill(bodyMiddle, downTrend, ST, (open + close) / 2, color.new(red, t1), color.new(red, t2))
plotshape(ta.crossunder(dir, 0) ? ST : na, "Bullish Trend", shape.labelup, location.absolute, green, text = "▲", textcolor = chart.fg_color, size = size.small) plotshape(ta.crossover(dir, 0) ? ST : na, "Bearish Trend", shape.labeldown, location.absolute, red, text = "▼", textcolor = chart.fg_color, size = size.small)
- Assistant
To improve your Pine Script, we'll address several key issues:
-
K-Means Implementation Errors: The current K-Means clustering implementation has logic errors that prevent it from functioning correctly. Specifically, the data points aren't being assigned exclusively to one cluster, and there's no handling for empty clusters.
-
Performance Optimization: The script can be optimized for better performance by reducing unnecessary calculations and iterations, which is crucial given Pine Script's resource limitations.
-
Code Readability and Maintainability: By refactoring and commenting on the code, we can make it more readable and easier to maintain.
Here's an improved version of your script with these considerations in mind:
//@version=5 strategy("Improved Machine Learning Adaptive SuperTrend", overlay = true, max_labels_count = 500) // Inputs atr_len = input.int(10, "ATR Length", group = "SuperTrend Settings") fact = input.float(3, "SuperTrend Factor", group = "SuperTrend Settings") training_data_period = input.int(100, "Training Data Length", group = "K-Means Settings") highvol = input.float(0.75, "Initial High Volatility Percentile Guess", maxval = 1, group = "K-Means Settings") midvol = input.float(0.5, "Initial Medium Volatility Percentile Guess", maxval = 1, group = "K-Means Settings") lowvol = input.float(0.25, "Initial Low Volatility Percentile Guess", maxval = 1, group = "K-Means Settings") green = input.color(#00ffbb, "Bullish Color", group = "Appearance") red = input.color(#ff1100, "Bearish Color", group = "Appearance") // Functions pine_supertrend(factor, atr) => src = hl2 upperBand = src + factor * atr lowerBand = src - 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 _dir = 1 _dir := nz(_dir[1]) _dir := _dir == -1 and close > upperBand ? 1 : _dir == 1 and close < lowerBand ? -1 : _dir superTrend = _dir == 1 ? lowerBand : upperBand [superTrend, _dir] // Main Code var float[] volatility_history = array.new_float() volatility = ta.atr(atr_len) array.unshift(volatility_history, volatility) // Run K-Means only after enough data is collected var float assigned_centroid = na if array.size(volatility_history) >= training_data_period // K-Means Clustering float[] data = array.slice(volatility_history, 0, training_data_period) // Initial centroids based on percentiles data_sorted = array.copy(data) array.sort(data_sorted) high_centroid = array.get(data_sorted, math.round(training_data_period * highvol) - 1) mid_centroid = array.get(data_sorted, math.round(training_data_period * midvol) - 1) low_centroid = array.get(data_sorted, math.round(training_data_period * lowvol) - 1) centroids = array.from(high_centroid, mid_centroid, low_centroid) prev_centroids = array.new_float(3) max_iterations = 10 iteration = 0 stop = false while not stop and iteration < max_iterations iteration += 1 // Create clusters clusters = array.new<array_float>(3) for i = 0 to 2 array.push(clusters, array.new_float()) // Assign data points to the nearest centroid for v in data dists = array.new_float() for c in centroids array.push(dists, math.abs(v - c)) min_dist = array.min(dists) cluster_idx = array.indexof(dists, min_dist) array.push(array.get(clusters, cluster_idx), v) // Update centroids stop = true for i = 0 to 2 cluster = array.get(clusters, i) if array.size(cluster) > 0 new_centroid = array.avg(cluster) if new_centroid != array.get(centroids, i) stop := false array.set(centroids, i, new_centroid) else // Handle empty cluster by keeping the centroid unchanged continue // Assign current volatility to the nearest centroid dists = array.new_float() for c in centroids array.push(dists, math.abs(volatility - c)) min_dist = array.min(dists) idx = array.indexof(dists, min_dist) assigned_centroid := array.get(centroids, idx) else assigned_centroid := volatility // Use current volatility if not enough data // Calculate SuperTrend [ST, dir] = pine_supertrend(fact, assigned_centroid) // Define buy and sell conditions longCondition = ta.crossunder(dir, 0) shortCondition = ta.crossover(dir, 0) // Execute buy and sell orders if (longCondition) strategy.entry("Long", strategy.long) if (shortCondition) strategy.entry("Short", strategy.short) // Stop Loss and Take Profit based on ATR atrVal = ta.atr(14) strategy.exit("Exit Long", from_entry = "Long", stop = close - 1.5 * atrVal, limit = close + 2 * atrVal) strategy.exit("Exit Short", from_entry = "Short", stop = close + 1.5 * atrVal, limit = close - 2 * atrVal) // Plotting t1 = 70 t2 = 95 upTrend = plot(close > ST ? ST : na, color = color.new(green, t1), style = plot.style_linebr) downTrend = plot(close < ST ? ST : na, color = color.new(red, t1), style = plot.style_linebr) bodyMiddle = plot(na, display = display.none) fill(bodyMiddle, upTrend, color.new(green, t2)) fill(bodyMiddle, downTrend, color.new(red, t2)) plotshape(ta.crossunder(dir, 0) ? ST : na, "Bullish Trend", shape.labelup, location.absolute, green, text = "▲", textcolor = color.white, size = size.small) plotshape(ta.crossover(dir, 0) ? ST : na, "Bearish Trend", shape.labeldown, location.absolute, red, text = "▼", textcolor = color.white, size = size.small)Explanation of Improvements:
-
Corrected the K-Means Clustering Logic:
- Exclusive Assignment: Each data point is now assigned to the nearest centroid exclusively.
- Handling Empty Clusters: If a cluster ends up empty after an iteration, the centroid remains unchanged to avoid division by zero errors.
- Convergence Criterion: The algorithm now stops when centroids no longer change or after a maximum number of iterations.
-
Optimized Performance:
- Limited Iterations: The maximum number of iterations for K-Means is set to 10 to prevent infinite loops.
- Data Handling: The script collects volatility data only when necessary and avoids unnecessary recalculations.
-
Enhanced Code Readability:
- Comments: Added comments throughout the script to explain each section.
- Structured Code Blocks: Organized the code into logical sections (inputs, functions, main code, plotting) for better readability.
- Simplified Variable Names: Used clear and concise variable names to make the code easier to understand.
-
Improved Plotting:
- Removed Unnecessary Plots: Simplified the plotting by removing redundant variables.
- Consistent Color Usage: Ensured consistent use of colors for bullish and bearish trends.
- Clear Trend Indicators: Used plotshape to mark bullish and bearish trend changes clearly.
Note: The K-Means algorithm in this script is a basic implementation and may not perform optimally in all market conditions. It's essential to test the strategy thoroughly and adjust the parameters (
training_data_period,max_iterations, initial centroid guesses) to fit the specific financial instrument and timeframe you are trading.Additional Recommendations:
- Backtesting: Before using this strategy in live trading, backtest it across different timeframes and market conditions to evaluate its performance.
- Risk Management: Adjust the stop-loss and take-profit levels according to your risk tolerance and the volatility of the instrument.
- Further Optimization: Consider experimenting with different clustering algorithms or machine learning techniques if you seek more advanced adaptability.
By addressing the core issues and optimizing the script, you should now have a more reliable and efficient strategy that adapts to market volatility using K-Means clustering.
-