我有这样的数据框:
| Date | Device | Status |
| 1990/01 | 50 | ON |
| 1990/01 | 20 | ON |
| 1990/03 | 25 | ON |
| 1990/05 | 50 | OFF |
| 2000/01 | 20 | OFF |
基本上,我会在发生设备状态时记录设备状态的更改,但这不是定期的,所以'日期'专栏不是连续的#39; 我想用任何简单的逻辑来查询设备状态,例如
df(device=50, date=1990/01) -> ON
- easy df(device=50, date=1990/02) -> ON
,设备50没有1990/02,但是当设备设置为特定状态时,除非没有变化,否则它将保留在其中df(device=50, date=1990/05) -> OFF
df(device=50, date=2000/09) -> OFF
df(device=50, date=1900/01) -> OFF
,这是一个棘手的问题,在第一次开启之前,设备处于关闭状态第二种情况具有不同的状态,但为了使问题简明扼要,我将保持不变。 除了第5点之外,行为完全相同。当我们假设首次登录状态之前的设备也处于ON状态时。 可能有更多的状态,但决定设备状态的逻辑完全相同。
如何以pandastic方式做到这一点?
答案 0 :(得分:1)
在我看来,必须在date_range
中定义所选的所有可能日期,用于reindex
个pivot
ed DataFrame的strftime
。首先将NaN
替换为NaN
的前向填充和OFF
的所有fillna
值:
print (df)
Date Device Status
0 1990/01 50 ON
1 1990/01 20 ON
2 1990/03 25 ON
3 1990/05 50 OFF
4 1990/05 20 OFF <-changed for smaller output df
df['Date'] = pd.to_datetime(df['Date'])
rng = pd.date_range('1989-10-01', '1991-01-01', freq='MS')
df = df.pivot('Date','Device','Status').reindex(rng).ffill().fillna('OFF')
print (df)
Device 20 25 50
1989-10-01 OFF OFF OFF
1989-11-01 OFF OFF OFF
1989-12-01 OFF OFF OFF
1990-01-01 ON OFF ON
1990-02-01 ON OFF ON
1990-03-01 ON ON ON
1990-04-01 ON ON ON
1990-05-01 OFF ON OFF
1990-06-01 OFF ON OFF
1990-07-01 OFF ON OFF
1990-08-01 OFF ON OFF
1990-09-01 OFF ON OFF
1990-10-01 OFF ON OFF
1990-11-01 OFF ON OFF
1990-12-01 OFF ON OFF
1991-01-01 OFF ON OFF
如果需要原始格式的dates
添加https://github.com/Smile-SA/elasticsuite/issues/912:
df.index = df.index.strftime('%Y/%m')
print (df)
Device 20 25 50
1989/10 OFF OFF OFF
1989/11 OFF OFF OFF
1989/12 OFF OFF OFF
1990/01 ON OFF ON
1990/02 ON OFF ON
1990/03 ON ON ON
1990/04 ON ON ON
1990/05 OFF ON OFF
1990/06 OFF ON OFF
1990/07 OFF ON OFF
1990/08 OFF ON OFF
1990/09 OFF ON OFF
1990/10 OFF ON OFF
1990/11 OFF ON OFF
1990/12 OFF ON OFF
1991/01 OFF ON OFF
编辑:
更一般的解决方案:
def get_status(df, device, check_date):
check_date = pd.to_datetime(check_date)
df['Date'] = pd.to_datetime(df['Date'])
rng = pd.date_range(df['Date'].min(), df['Date'].max(), freq='MS')
df = df.pivot('Date','Device','Status').reindex(rng).ffill().fillna('OFF')
#print (df)
if check_date < df.index.min():
return 'OFF'
elif check_date > df.index.max():
return df.loc[df.index[-1], device]
else:
return df.loc[check_date, device]
print (get_status(df, 50, '1990/01'))
#ON
print (get_status(df, 50, '1990/02'))
#ON
print (get_status(df, 50, '1990/05'))
#OFF
print (get_status(df, 50, '1990/09'))
#OFF
print (get_status(df, 50, '1900/01'))
#OFF
答案 1 :(得分:1)
假设您的原始数据框名为status_df
,并且您已将Date
列转换为datetime.date
类型。然后,您可以定义一个函数并使用status_df
作为第一个参数和所需的device
和check_date
def get_status(df, device, check_date):
# Filter only those entries for this device and with date <= check_date
filtered_df = df[(df['Device']==device) & (df['Date']<=check_date)]
# Now sort by date
filtered_df.sort_values('Date', inplace=True)
# Now the last entry of the filtered_df contains the active status at check_date
# If it is empty choose a default values (say OFF)
if len(filtered_df) < 1:
return 'OFF'
return filtered_df.iloc[-1]['Status']
<强>更新强> 如果您不想创建新的数据框,可以使用简单的try语句来完成
try:
return df[(df['Device']==device) & (df['Date']<=check_date)].sort_values('Date').iloc[-1]['Status']
except IndexError:
return 'OFF'