Question

我有一个带有时间戳和两列的日志文件。我现在想重新取样和＃34; pivot＆＃34;从日志文件创建的日期框架。

示例orig数据帧/日志文件：

timestamp            colA  colB

2015-01-01 00:10:01   a     x
2014-01-01 00:10:01   b     y
2015-01-01 00:10:03   a     x
2015-01-01 00:10:03   a     x
2015-01-01 00:10:03   a     y
2015-01-01 00:10:04   b     x
2014-01-01 00:10:04   b     y
2014-01-01 00:10:04   b     y
2014-01-01 00:10:04   a     x
2014-01-01 00:10:05   a     x
2014-01-01 00:10:05   a     x
2014-01-01 00:10:07   a     y
2014-01-01 00:10:08   a     x

按秒重新采样的示例结果：

                      a           b
timestamp             x     y     x     y

2015-01-01 00:10:01   1     0     0     1
2015-01-01 00:10:02   0     0     0     0
2015-01-01 00:10:03   2     1     0     0
2015-01-01 00:10:04   1     0     1     2
2014-01-01 00:10:05   2     0     0     0
2014-01-01 00:10:06   0     0     0     0
2014-01-01 00:10:07   0     1     0     0
2014-01-01 00:10:08   1     0     0     0

我将如何实现这一目标？首先重新采样，然后是groupby / pivot？或者相反？更具体地说，单元格应该包含每个特定重采样时间间隔的colA / colB组合的计数。在示例秒中，但可能是分钟，小时等。

我不是固定在这种格式上，我也可以考虑获得重新采样的结果和groupby timestamp / colA之类的

                             colB
timestamp              colA   x     y

2015-01-01 00:10:01     a     1     0
                        b     0     1
2015-01-01 00:10:02     a     0     0
                        b     0     0
2015-01-01 00:10:03     a     2     1
                        b     0     0
2015-01-01 00:10:04     a     1     0
                        b     1     2
2014-01-01 00:10:05     a     2     0
                        b     0     0
2014-01-01 00:10:06     a     0     0
                        b     0     0
2014-01-01 00:10:07     a     0     1
                        b     0     0
2014-01-01 00:10:08     a     1     0
                        b     0     0

最终用法是绘制不同的计数值

THX。

Answer 1

您可以使用pd.crosstab：

import numpy as np
import pandas as pd

df = pd.read_table('data', sep='\s{2,}', parse_dates=[0])
table = pd.crosstab(index=[df['timestamp']], columns=[df['colA'], df['colB']])

产量

colA                 a     b   
colB                 x  y  x  y
timestamp                      
2014-01-01 00:10:01  0  0  0  1
2014-01-01 00:10:04  1  0  0  2
2014-01-01 00:10:05  2  0  0  0
2014-01-01 00:10:07  0  1  0  0
2014-01-01 00:10:08  1  0  0  0
2015-01-01 00:10:01  1  0  0  0
2015-01-01 00:10:03  2  1  0  0
2015-01-01 00:10:04  0  0  1  0

重新采样，分组，转动pandas数据帧

1 个答案: