如何在Python中按天匯總時間序列數據？ resample.sum（）無效

Question

我是Python的新手。 如何根據日期對數據求和並繪制結果？

我有一個帶有數據的Series對象：

2017-11-03 07:30:00      NaN
2017-11-03 09:18:00      NaN
2017-11-03 10:00:00      NaN
2017-11-03 11:08:00      NaN
2017-11-03 14:39:00      NaN
2017-11-03 14:53:00      NaN
2017-11-03 15:00:00      NaN
2017-11-03 16:00:00      NaN
2017-11-03 17:03:00      NaN
2017-11-03 17:42:00    800.0
2017-11-04 07:27:00    600.0
2017-11-04 10:10:00      NaN
2017-11-04 11:48:00      NaN
2017-11-04 12:58:00    500.0
2017-11-04 13:40:00      NaN
2017-11-04 15:15:00      NaN
2017-11-04 16:21:00      NaN
2017-11-04 17:37:00    500.0
2017-11-04 21:37:00      NaN
2017-11-05 03:00:00      NaN
2017-11-05 06:30:00      NaN
2017-11-05 07:19:00      NaN
2017-11-05 08:31:00    200.0
2017-11-05 09:31:00    500.0
2017-11-05 12:03:00      NaN
2017-11-05 12:25:00    200.0
2017-11-05 13:11:00    500.0
2017-11-05 16:31:00      NaN
2017-11-05 19:00:00    500.0
2017-11-06 08:08:00      NaN

我有以下代碼：

# load packages
import pandas as pd
import matplotlib.pyplot as plt

# import painkiller data
df = pd.read_csv('/Users/user/Documents/health/PainOverTime.csv',delimiter=',')

# plot bar graph of date and painkiller amount
times = pd.to_datetime(df.loc[:,'Time'])

ts = pd.Series(df.loc[:,'acetaminophen'].values, index = times,
               name = 'Painkiller over Time')
ts.plot()

這給了我下面的line（？）圖：

這是一個開始； 現在我想按日期對劑量求和。 但是，此代碼無法實現任何更改：結果圖相同。 怎么了？

ts.resample('D',closed='left', label='right').sum()
ts.plot()

我也嘗試過ts.resample('D').sum() ， ts.resample('1d').sum() ， ts.resample('1D').sum() ，但是情節。

.resample甚至是正確的功能嗎？ 我理解重采樣是從數據中采樣 ，例如，每天隨機抽取一個點，而我想將每天的值相加。

即，我希望有一些結果（基於上述數據），例如：

2017-11-03 800
2017-11-04 1600
2017-11-05 1900
2017-11-06 NaN

Answer 1

使用pandas groupby函數。

import io
import pandas as pd

data = io.StringIO('''
2017-11-03 07:30:00,NaN
2017-11-03 09:18:00,NaN
2017-11-03 10:00:00,NaN
2017-11-03 11:08:00,NaN
2017-11-03 14:39:00,NaN
2017-11-03 14:53:00,NaN
2017-11-03 15:00:00,NaN
2017-11-03 16:00:00,NaN
2017-11-03 17:03:00,NaN
2017-11-03 17:42:00,800.0
2017-11-04 07:27:00,600.0
2017-11-04 10:10:00,NaN
2017-11-04 11:48:00,NaN
2017-11-04 12:58:00,500.0
2017-11-04 13:40:00,NaN
2017-11-04 15:15:00,NaN
2017-11-04 16:21:00,NaN
2017-11-04 17:37:00,500.0
2017-11-04 21:37:00,NaN
2017-11-05 03:00:00,NaN
2017-11-05 06:30:00,NaN
2017-11-05 07:19:00,NaN
2017-11-05 08:31:00,200.0
2017-11-05 09:31:00,500.0
2017-11-05 12:03:00,NaN
2017-11-05 12:25:00,200.0
2017-11-05 13:11:00,500.0
2017-11-05 16:31:00,NaN
2017-11-05 19:00:00,500.0
2017-11-06 08:08:00,NaN
''')
column_names = ['date', 'val']
df = pd.read_csv(data, sep=',', header = None, names = column_names)
df['date'] = pd.to_datetime(df['date'])
df = df.groupby(df['date'].dt.date)[['val']].sum()
df.plot()

Answer 2

這個答案幫助我了解到我需要將其分配給新對象（如果這是正確的術語）：

import pandas as pd
import matplotlib.pyplot as plt
df = pd.read_csv('/Users/user/Documents/health/PainOverTime.csv',delimiter=',')
# plot bar graph of date and painkiller amount
times = pd.to_datetime(df.loc[:,'Time'])

# raw plot of data
ts = pd.Series(df.loc[:,'acetaminophen'].values, index = times,
               name = 'Painkiller over Time')
fig1 = ts.plot()

# combine data by day
test2 = ts.resample('D').sum()
fig2 = test2.plot()

這將產生以下圖：

這種方法是否不比“ groupby”功能更好？

現在如何制作散點圖或條形圖而不是此線圖...？

Answer 3

簡短答案：您需要.groupby() ，而不是.resample() ，如本答案所示

較長的代碼：

import pandas as pd
from io import StringIO

doc = StringIO("""2017-11-03 07:30:00      NaN
2017-11-03 09:18:00      NaN
2017-11-03 10:00:00      NaN
2017-11-03 11:08:00      NaN
2017-11-03 14:39:00      NaN
2017-11-03 14:53:00      NaN
2017-11-03 15:00:00      NaN
2017-11-03 16:00:00      NaN
2017-11-03 17:03:00      NaN
2017-11-03 17:42:00    800.0
2017-11-04 07:27:00    600.0
2017-11-04 10:10:00      NaN
2017-11-04 11:48:00      NaN
2017-11-04 12:58:00    500.0
2017-11-04 13:40:00      NaN
2017-11-04 15:15:00      NaN
2017-11-04 16:21:00      NaN
2017-11-04 17:37:00    500.0
2017-11-04 21:37:00      NaN
2017-11-05 03:00:00      NaN
2017-11-05 06:30:00      NaN
2017-11-05 07:19:00      NaN
2017-11-05 08:31:00    200.0
2017-11-05 09:31:00    500.0
2017-11-05 12:03:00      NaN
2017-11-05 12:25:00    200.0
2017-11-05 13:11:00    500.0
2017-11-05 16:31:00      NaN
2017-11-05 19:00:00    500.0
2017-11-06 08:08:00      NaN""")

df = pd.read_csv(doc, sep='\\s{2,}', 
                 header=None, 
                 converters={'timestamp': pd.to_datetime}, 
                 names = ['timestamp', 'acetaminophen'],
                 engine='python')
df = df.set_index('timestamp')

#true, but rather ugly x axis line
df.plot.bar()

df1 = df.groupby(by=[df.index.date]).sum()
df1.plot.bar()

如果日期不是連續的，則可以創建一個具有完整timeindex的空數據框，並將df1與之合並。