np.select 而不是 for while 循环答案

【问题标题】：np.select instead of for while loopnp.select 而不是 for while 循环
【发布时间】：2020-12-10 18:45:19
【问题描述】：

我的目标是显着加快我的代码速度，我认为可以使用 np.select 来完成，尽管我不知道如何。

这是我的代码执行时的当前输出：

date  starting_temp  average_high  average_low  limit_temp observation_date   Date_Limit_reached
2019-12-03 22:30:00 NaN             13.0          14.8        NaN                          nan                
2019-12-03 23:00:00 NaN             14.7          14.9        NaN                          nan                
2019-12-03 23:30:00 NaN             13.0          13.9        NaN                          nan                
2019-12-04 00:00:00  13.2           13.0          14.7        NaN                          2019-12-04 10:00:00
2019-12-04 00:30:00 NaN             14.0          13.8        NaN                          nan                
2019-12-04 01:00:00 NaN             13.9          13.8        NaN                          nan                
2019-12-04 01:30:00 NaN             13.6          14.8        NaN                          nan                
2019-12-04 02:00:00 NaN             13.1          14.5        NaN                          nan                
2019-12-04 02:30:00 NaN             14.9          13.7        NaN                          nan                
2019-12-04 03:00:00 NaN             14.2          14.1        NaN                          nan                
2019-12-04 03:30:00 NaN             13.4          14.1        NaN                          nan                
2019-12-04 04:00:00 NaN             14.3          13.0        NaN                          nan                
2019-12-04 04:30:00 NaN             13.5          14.1        NaN                          nan                
2019-12-04 05:00:00 NaN             13.6          13.4        NaN                          nan                
2019-12-04 05:30:00 NaN             14.5          13.9        NaN                          nan                
2019-12-04 06:00:00 NaN             14.4          14.5        NaN                          nan                
2019-12-04 06:30:00 NaN             13.7          14.2        NaN                          nan                
2019-12-04 07:00:00 NaN             13.7          14.2        NaN                          nan                
2019-12-04 07:30:00 NaN             13.2          14.4        NaN                          nan                
2019-12-04 08:00:00 NaN             13.9          13.1        NaN                          nan                
2019-12-04 08:30:00 NaN             13.9          14.4        NaN                          nan                
2019-12-04 09:00:00 NaN             14.4          13.9        NaN                          nan                
2019-12-04 09:30:00 NaN             14.4          13.8        NaN                          nan                
2019-12-04 10:00:00 NaN             15.0          14.0        NaN                          nan                
2019-12-04 10:30:00 NaN             13.2          13.2        NaN                          nan                
2019-12-04 11:00:00 NaN             14.0          13.3        NaN                          nan                
2019-12-04 11:30:00 NaN             14.2          13.4        NaN                          nan                
2019-12-04 12:00:00 NaN             14.2          13.4        NaN                          nan                
2019-12-04 12:30:00 NaN             13.7          13.6        NaN                          nan                
2019-12-04 13:00:00 NaN             14.1          13.3        NaN                          nan                
2019-12-04 13:30:00 NaN             13.1          14.1        NaN                          nan                
2019-12-04 14:00:00 NaN             13.2          14.3        NaN                          nan                
2019-12-04 14:30:00 NaN             13.7          13.8        NaN                          nan

生成最终 df['Date_Limit_reached'] 列的代码太慢了，我在下面添加了。如果可能，我想将其结构更改为np.select：

    new_col = []
    
    df_size = len(df)
    
    # Loop the dataframe
    for ind in df.index:
        if not math.isnan(df['starting_temp'][ind]):   
            entry_price_val = df['starting_temp'][ind]
            count = 0
            hasValue = False
    
            while count < df_size:
       
                if df['starting_temp'][ind] > df['limit_temp'][ind] and df['limit_temp'][ind] >= df['asklow'][count] and df['date'][count] >= df['observation_date'][ind] :
                    new_col.append(df['date'][count])
                    hasValue = True
                    break  # Break the loop if matching value meets
                    count += 1
    
                elif df['starting_temp'][ind] < df['limit_temp'][ind] and df['limit_temp'][ind] <= df['average_high'][count] and df['date'][count] >= df['observation_date'][ind] :
                    new_col.append(df['date'][count])
                    hasValue = True
                    break  # Break the loop if matching value meets
                count += 1            
    
            # If matching value not meets, then append nan value to the column   
            if not hasValue:
                new_col.append(float('nan'))
        else:
            new_col.append(float('nan'))
    
 
   df['Date_Limit_reached'] = new_col

【问题讨论】：

什么是df？代码中没有定义。
您能否提供一个代码来创建示例数据框df，以便我们可以轻松地就列类型达成一致、测试您的代码并讨论类似数据的性能？

标签： python performance numpy for-loop

【解决方案1】：

由于此处缺少 df 我无法运行代码，因此我的建议是：

使用更少的标志，而是使用具体的值。使代码更具可读性。 hasValue --> 值
如果有df['starting_temp'][ind] == df['limit_temp'][ind] 的条目，您将遇到问题，因为您的任何案例都不会触发。也许这就是代码慢的问题。
您可以预先计算 while 循环中的第一个布尔表达式。这可能会解决上述问题
你不使用entry_price_val
为了进一步改进，请使用数据矢量化，这在所有循环中都是可能的。（我的代码中没有显示，因为我无法测试它）

这是我建议的代码

new_col = []
df_size = len(df)    
for ind in df.index:
    val = float('nan') # use data instead of flags
    
    if not math.isnan(df['starting_temp'][ind]):   
        count = 0
        
        if df['starting_temp'][ind] > df['limit_temp'][ind]:
            while count < df_size:
                if df['limit_temp'][ind] >= df['asklow'][count] and df['date'][count] >= df['observation_date'][ind] :
                    val=df['date'][count]
                    break  # Break the loop if matching value meets
                count += 1  
        elif df['starting_temp'][ind] < df['limit_temp'][ind]
            while count < df_size:
                if df['limit_temp'][ind] <= df['average_high'][count] and df['date'][count] >= df['observation_date'][ind] :
                    val = df['date'][count]
                    break  # Break the loop if matching value meets
                count += 1            
    new_col.append(val)
df['Date_Limit_reached'] = new_col

代码 sn-ps 未经测试，需要测试正确性，可能会进一步改进（根据要求提供提示）。

【讨论】：