计算持仓成本，有没有高效的方法

Question

如下代码所示，order表的字段id是订单号，price是价格，amount下单数量（+买入，-卖出），当amount为正时，持仓成本=cumsum(amount*price)/cumsum(amount)，当amount为负（卖出时），引用之前计算好的持仓成本。当全卖出即空仓后，持仓成本需要重新算。下面是我的DolphinDB database代码：

id = take(1..9,1000000)
price = take(101..109,1000000)
amount =take( 1 2 3 -2 -1 -3 1 2 3,1000000)
order = table(id, price, amount)

def avg_price(price, amount){
 hold = 0.0;
 cost = 0.0;
 avg_price = 0.0;
 avg_prices = array(double);
 for ( r in table(price,amount) ) {
  hold += r.amount;
  if ( r.amount > 0 ) {
   cost += r.amount * r.price;
   avg_price = iif(hold > 0, cost/hold, 0);
  } else {
   cost += r.amount * avg_price;
  }
  avg_prices.push!(avg_price)
 }
 return avg_prices;
}

select *, iif (amount < 0, amount*(avg_price - price), 0) as profit 
from (
 select *, avg_price(price, amount) as avg_price
 from order
) 

这个方法效率比较低，100万行数据，执行一次要3-4秒，请问有什么办法改进？

Answer 1

DolphinDB database 从1.01版本开始支持即时编译JIT（https://github.com/dolphindb/... ）。JIT又译及时编译或实时编译，是动态编译的一种形式，可提高程序运行效率。上面函数可写成JIT函数如下：

@jit
def avg_price(price, amount){
    hold = 0.0;
    cost = 0.0;
    avgPrice = 0.0;
    n = size(price)
    avgPrices = array(DOUBLE, n, n, 0)
    for ( i in 0..(n-1))  {
        hold =hold + amount[i];
        if ( amount[i] > 0 ) {
            cost = cost + amount[i] * price[i];
            if(hold > 0) {
                avgPrice = cost/hold;
            }else{  
                avgPrice = 0.0;
            }
        } else {
             cost += amount[i] * avgPrice;; 
        }
        avgPrices[i]=avgPrice
    }
    return avgPrices;
}

对上述函数进行测试，用JIT版本比非JIT大约快200倍。