FMZ量化交易手把手教你写一个Python版的K线合成函数

手把手教你写一个Python版的K线合成函数

在编写、使用策略时，经常会使用一些不常用的K线周期数据。然而交易所、数据源又没有提供这些周期的数据。只能通过使用已有周期的数据进行合成。合成算法已经有一个JavaScript版本了（链接），其实移植一段JavaScript代码为Python版本很简单。接下来我们一起写一个Python版本的K线合成算法。

JavaScript版本

  function GetNewCycleRecords (sourceRecords, targetCycle) {    // K线合成函数
      var ret = []

      // 首先获取源K线数据的周期
      if (!sourceRecords || sourceRecords.length < 2) {
          return null
      }
      var sourceLen = sourceRecords.length
      var sourceCycle = sourceRecords[sourceLen - 1].Time - sourceRecords[sourceLen - 2].Time

      if (targetCycle % sourceCycle != 0) {
          Log("targetCycle:", targetCycle)
          Log("sourceCycle:", sourceCycle)
          throw "targetCycle is not an integral multiple of sourceCycle."
      }

      if ((1000 * 60 * 60) % targetCycle != 0 && (1000 * 60 * 60 * 24) % targetCycle != 0) {
          Log("targetCycle:", targetCycle)
          Log("sourceCycle:", sourceCycle)
          Log((1000 * 60 * 60) % targetCycle, (1000 * 60 * 60 * 24) % targetCycle)
          throw "targetCycle cannot complete the cycle."
      }

      var multiple = targetCycle / sourceCycle


      var isBegin = false
      var count = 0
      var high = 0
      var low = 0
      var open = 0
      var close = 0
      var time = 0
      var vol = 0
      for (var i = 0 ; i < sourceLen ; i++) {
          // 获取 时区偏移数值
          var d = new Date()
          var n = d.getTimezoneOffset()

          if (((1000 * 60 * 60 * 24) - sourceRecords[i].Time % (1000 * 60 * 60 * 24) + (n * 1000 * 60)) % targetCycle == 0) {
              isBegin = true
          }

          if (isBegin) {
              if (count == 0) {
                  high = sourceRecords[i].High
                  low = sourceRecords[i].Low
                  open = sourceRecords[i].Open
                  close = sourceRecords[i].Close
                  time = sourceRecords[i].Time
                  vol = sourceRecords[i].Volume

                  count++
              } else if (count < multiple) {
                  high = Math.max(high, sourceRecords[i].High)
                  low = Math.min(low, sourceRecords[i].Low)
                  close = sourceRecords[i].Close
                  vol += sourceRecords[i].Volume

                  count++
              }

              if (count == multiple || i == sourceLen - 1) {
                  ret.push({
                      High : high,
                      Low : low,
                      Open : open,
                      Close : close,
                      Time : time,
                      Volume : vol,
                  })
                  count = 0
              }
          }
      }

      return ret
  }

有JavaScript算法，对于Python其实逐行翻译移植就可以了，遇到JavaScript的内置函数，或者固有方法，对应的去Python中查找对应的方法即可，所以移植还是比较容易的。
算法逻辑完全一模一样，只是JavaScript的函数调用var n = d.getTimezoneOffset()，移植到Python时，使用Python的time库中的n = time.altzone代替。其它差异仅仅是语言语法方面的了（例如for循环的使用，布尔值的差别，逻辑与、逻辑非、逻辑或的使用差别等..）。

移植后的Python代码：

import time

def GetNewCycleRecords(sourceRecords, targetCycle):
    ret = []

    # 首先获取源K线数据的周期
    if not sourceRecords or len(sourceRecords) < 2 :
        return None

    sourceLen = len(sourceRecords)
    sourceCycle = sourceRecords[-1]["Time"] - sourceRecords[-2]["Time"]

    if targetCycle % sourceCycle != 0 :
        Log("targetCycle:", targetCycle)
        Log("sourceCycle:", sourceCycle)
        raise "targetCycle is not an integral multiple of sourceCycle."

    if (1000 * 60 * 60) % targetCycle != 0 and (1000 * 60 * 60 * 24) % targetCycle != 0 :
        Log("targetCycle:", targetCycle)
        Log("sourceCycle:", sourceCycle)
        Log((1000 * 60 * 60) % targetCycle, (1000 * 60 * 60 * 24) % targetCycle)
        raise "targetCycle cannot complete the cycle."

    multiple = targetCycle / sourceCycle

    isBegin = False
    count = 0
    barHigh = 0
    barLow = 0
    barOpen = 0
    barClose = 0
    barTime = 0
    barVol = 0

    for i in range(sourceLen) :
        # 获取时区偏移数值
        n = time.altzone

        if ((1000 * 60 * 60 * 24) - (sourceRecords[i]["Time"] * 1000) % (1000 * 60 * 60 * 24) + (n * 1000)) % targetCycle == 0 :
            isBegin = True

        if isBegin :
            if count == 0 :
                barHigh = sourceRecords[i]["High"]
                barLow = sourceRecords[i]["Low"]
                barOpen = sourceRecords[i]["Open"]
                barClose = sourceRecords[i]["Close"]
                barTime = sourceRecords[i]["Time"]
                barVol = sourceRecords[i]["Volume"]
                count += 1
            elif count < multiple :
                barHigh = max(barHigh, sourceRecords[i]["High"])
                barLow = min(barLow, sourceRecords[i]["Low"])
                barClose = sourceRecords[i]["Close"]
                barVol += sourceRecords[i]["Volume"]
                count += 1

            if count == multiple or i == sourceLen - 1 :
                ret.append({
                    "High" : barHigh,
                    "Low" : barLow,
                    "Open" : barOpen,
                    "Close" : barClose,
                    "Time" : barTime,
                    "Volume" : barVol,
                })
                count = 0

    return ret

# 测试
def main():
    while True:
        r = exchange.GetRecords()
        r2 = GetNewCycleRecords(r, 1000 * 60 * 60 * 4)

        ext.PlotRecords(r2, "r2")
        Sleep(1000)

测试

火必行情图表

回测合成4小时图表

以上代码仅作为学习参考使用，如果用于具体策略中，请根据需求修改、测试。
如有BUG或者改进建议，欢迎留言，十分感谢 o^_^o