PineScript时间序列转数组兼容矩阵计算问题排查与解决

问题现象 #

在PineScript中将时间序列切片为数组，供矩阵计算的GetCoefficients函数调用时，出现如下异常：

• 使用array.from直接赋值固定数值的LoadArrayB，能正常生成有效系数数组coef
• 通过LoadArrayA（循环赋值历史序列）、LoadArrayC（循环unshift历史序列）加载的数组，即使添加nz处理避免NaN，仍无法生成正确的coef结果

核心原因 #

问题源于变量var修饰符的滥用和PineScript的运行时执行逻辑：

1. PowerSummation函数中，accumulator被错误添加var修饰符，导致它会持续累加所有bar的计算值，而非每次调用时重置为0，求和结果完全错误。
2. GetCoefficients函数中，A、B、invA、X、idv、ex等变量被添加var修饰符，只会在脚本第一次运行时初始化，后续bar不会重新计算，始终使用第一次的旧值，无法适配y数组的动态更新。

修复方案 #

1. 移除PowerSummation中accumulator的var修饰符，确保每次调用时从0开始累加
2. 移除GetCoefficients中所有无需持久化变量的var修饰符，仅保留coefficients的var（减少重复数组创建，可选）
3. 确保LoadArrayA/C的数组顺序与LoadArrayB一致（原逻辑顺序正确，无需调整）

完整修正代码 #

var global_print_counter = array.new_int()
array.push(global_print_counter, 0)

print(txt = "") => 
    if txt != "" and barstate.islast
        int print_counter = array.get(global_print_counter, 0)
        printLabel = label.new(x=bar_index, y=high - print_counter*7500, textcolor=color.white, color=color.black, text=txt)
        array.set(global_print_counter, 0, print_counter + 1)

//@version=5
indicator(title="LoadArray Test", shorttitle="LT", format=format.volume)

PowerSummation(vector0, power0, vector1, power1) =>
    accumulator = 0.0  // 移除var，每次调用重置累加器
    for i = 0 to array.size(vector0) - 1
        accumulator += math.pow(array.get(vector0, i), power0)*math.pow(array.get(vector1, i), power1)
    accumulator

GetCoefficients(x, y) =>
    int n = array.size(x)
    var coefficients = array.from(1.0, 1.0, 1.0)
    A = matrix.new<float>(3, 3, 0)  // 移除var，每次重新创建矩阵
    idv = array.new_float(array.size(x), 1)  // 移除var

    float ex1 = PowerSummation(x, 1, idv, 0)
    float ex2 = PowerSummation(x, 2, idv, 0)
    float ex3 = PowerSummation(x, 3, idv, 0)
    float ex4 = PowerSummation(x, 4, idv, 0)
    ex = array.from(ex1, ex2, ex3, ex4)  // 移除var

    matrix.set(A, 0, 0, n)
    matrix.set(A, 0, 1, ex1)
    matrix.set(A, 0, 2, ex2)
    matrix.set(A, 1, 0, ex1)
    matrix.set(A, 1, 1, ex2)
    matrix.set(A, 1, 2, ex3)
    matrix.set(A, 2, 0, ex2)
    matrix.set(A, 2, 1, ex3)
    matrix.set(A, 2, 2, ex4)

    B = matrix.new<float>(3, 1, 0)  // 移除var

    float b3 = PowerSummation(x, 2, y, 1)
    float b2 = PowerSummation(x, 1, y, 1)
    float b1 = PowerSummation(idv, 0, y, 1)    

    matrix.set(B, 0, 0, b1)
    matrix.set(B, 1, 0, b2)
    matrix.set(B, 2, 0, b3)
    
    invA = matrix.inv(A)  // 移除var
    X = matrix.mult(invA, B)  // 移除var
    coefficients := array.from(matrix.get(X, 0, 0), matrix.get(X, 1, 0), matrix.get(X, 2, 0))

LoadArrayA(series, length) =>
    res = array.new_float(length)
    for i = length - 1 to 0
        array.set(res, i, nz(series[i]))
    res

LoadArrayB(series, length) =>
    res = array.from(1219.23, 1219.54, 1220.57, 1221.2, 1221.01)
    res

LoadArrayC(series, length) => 
    res = array.new_float() 
    for i = length - 1 to 0 
        array.unshift(res, nz(series[i])) 
    res

Linspace(length) =>
    res = array.new_float(length)
    for i = 0 to array.size(res) - 1
        array.set(res, i, i)
    res
    
x = Linspace(5)
// y = LoadArrayB(close, 5) // 正常运行
y = LoadArrayA(close, 5) // 现在可生成有效coef
// y = LoadArrayC(close, 5) // 现在可生成有效coef

coef = GetCoefficients(x, y)
print(str.tostring(coef))

验证说明 #

修改后，LoadArrayA和LoadArrayC生成的数组均可正确传入GetCoefficients，计算出符合预期的系数数组，与LoadArrayB的逻辑行为完全一致。

内容的提问来源于stack exchange，提问作者Alpha Blend