Question

尝试iterrows（）非常慢，在别处读取zip会更好，但它仍然很慢。

我尝试搜索数据帧的行，生成一些统计信息以填充两个新的数据帧。

是否有任何建议加快搜索数据帧的行？

代码段：

for index,date,stocknum in zip(stockpicks.index.values,stockpicks.date.values,stockpicks.stocknum.values):

        stock=readStockPrice(stocknum)
        if stock.empty:
            return print("error - empty frame")
        stock=stock.ix[trading_days]
        stockprice=stock.Close.values
        p0_date=trading_days.get_loc(date)


        p0=stockprice[p0_date]
        stock_pct_change={('d'+str(d)):stockprice[p0_date+d]/p0*100.0 if (p0_date+d)< len(trading_days) else np.nan for d in days }

        b0=hsi[p0_date]
        benchmark_pct_change={('d'+str(d)):hsi[p0_date+d]/b0*100.0 if (p0_date+d)< len(trading_days) else np.nan for d in days }

        for d in days:
            stock_analysis.loc[index,'d'+str(d)]=stock_pct_change['d'+str(d)]
            benchmark_analysis.loc[index,'d'+str(d)]=benchmark_pct_change['d'+str(d)]

Answer 1

您出现的问题可以完全矢量化。像你一样进行迭代和索引是最慢的方法。

In [6]: df = DataFrame(np.random.randint(-5,5,size=20).reshape(5,4),columns=list('abcd'),index=date_range('20130101',periods=5))+50.0

In [7]: df.pct_change()
Out[7]: 
                   a         b         c         d
2013-01-01       NaN       NaN       NaN       NaN
2013-01-02  0.108696  0.108696  0.102041  0.086957
2013-01-03 -0.058824 -0.039216 -0.074074 -0.060000
2013-01-04  0.104167  0.081633 -0.020000  0.000000
2013-01-05 -0.075472 -0.113208  0.061224 -0.021277

[5 rows x 4 columns]

如何加速数据帧行的迭代

1 个答案: