您的位置: 首页  >  IT文章  >  用cython提升python的性能

用cython提升python的性能

分类: IT文章 • 2023-11-24 13:44:25

Boosting performance with Cython

 
 
Even with my old pc (AMD Athlon II, 3GB ram), I seldom run into performance issues when running vectorized code. But unfortunately there are plenty of cases where that can not be easily vectorized, for example the drawdown function. My implementation of such was extremely slow, so I decided to use it as a test case for speeding things up. I'll be using the SPY timeseries with ~5k samples as test data. Here comes the original version of my drawdown function (as it is now implemented in the TradingWithPython library) 
?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
def drawdown(pnl):
    """
    calculate max drawdown and duration
 
    Returns:
        drawdown : vector of drawdwon values
        duration : vector of drawdown duration
    """
    cumret = pnl
 
    highwatermark = [0]
 
    idx = pnl.index
    drawdown = pd.Series(index = idx)
    drawdowndur = pd.Series(index = idx)
 
    for t in range(1, len(idx)) :
        highwatermark.append(max(highwatermark[t-1], cumret[t]))
        drawdown[t]= (highwatermark[t]-cumret[t])
        drawdowndur[t]= (0 if drawdown[t] == 0 else drawdowndur[t-1]+1)
 
    return drawdown, drawdowndur
 
%timeit drawdown(spy)
1 loops, best of 3: 1.21 s per loop
Hmm 1.2 seconds is not too speedy for such a simple function. There are some things here that could be a great drag to performance, such as a list *highwatermark* that is being appended on each loop iteration. Accessing Series by their index should also involve some processing that is not strictly necesarry. Let's take a look at what happens when this function is rewritten to work with numpy data 
?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
def dd(s):
#    ''' simple drawdown function '''
     
    highwatermark = np.zeros(len(s))
    drawdown = np.zeros(len(s))
    drawdowndur = np.zeros(len(s))
 
  
    for t in range(1,len(s)):
        highwatermark[t] = max(highwatermark[t-1], s[t])
        drawdown[t] = (highwatermark[t]-s[t])
        drawdowndur[t]= (0 if drawdown[t] == 0 else drawdowndur[t-1]+1)
        
      
    return drawdown , drawdowndur
 
%timeit dd(spy.values)
10 loops, best of 3: 27.9 ms per loop
Well, this is much faster than the original function, approximately 40x speed increase. Still there is much room for improvement by moving to compiled code with cython Now I rewrite the dd function from above, but using optimisation tips that I've found on the cython tutorial .

相关推荐