Question

根据问题标题。熊猫可以依靠RollingGroupby对象上的字符串型列吗？

这是我的数据框：

# Let's say my objective is to count the number of unique cars 
# over the last 1 day grouped by park

 park |    date    | to_count
------------------------------
  A   | 2019-01-01 |   Honda
  A   | 2019-01-03 |   Lexus
  A   | 2019-01-05 |   BMW
  A   | 2019-01-05 |   Lexus
  B   | 2019-01-01 |   BMW
  B   | 2019-01-08 |   Lexus
  B   | 2019-01-08 |   Lexus
  B   | 2019-01-10 |   Ford

这就是我想要的：

 park |    date    | unique_count
----------------------------------
  A   | 2019-01-01 |      1
  A   | 2019-01-03 |      1
  A   | 2019-01-05 |      2
  B   | 2019-01-01 |      1
  B   | 2019-01-08 |      1
  B   | 2019-01-10 |      1

# Bit of explanation:
# There are 2 type of cars coming to park A over last 1 day on 5th Jan so distinct count is 2.
# There are 2 cars of 1 type (Lexus) coming to park B over last 1 day on 8th Jan so distinct count is 1.

这是我尝试过的：

import pandas as pd
import numpy as np

# initiate dataframe
df = pd.DataFrame({
    'park': ['A', 'A', 'A', 'A', 'B', 'B', 'B', 'B'],
    'date': ['2019-01-01', '2019-01-03', '2019-01-05', '2019-01-05',
             '2019-01-01', '2019-01-08', '2019-01-08', '2019-01-10'],
    'to_count': ['Honda', 'Lexus', 'BMW', 'Lexus', 'BMW', 'Lexus', 'Lexus', 'Ford']
})

# string to date
df['date'] = pd.to_datetime(df['date'])

# group. This is more intuitive to me but sadly this does not work.
unique_count = df.groupby('park').rolling('1d', on='date').to_count.nunique()

# factorize then group. This works (but why???)
df['factorized'] = pd.factorize(df.to_count)[0]
unique_count = df.groupby('park').rolling('1d', on='date').factorized.apply(lambda x: len(np.unique(x)) )

result = unique_count.reset_index().drop_duplicates(subset=['park', 'date'], keep='last')

这是我的环境：

Mac 10.12 High Sierra
python3.6
熊猫0.22.0

为了强调，我需要滚动窗口功能才能正常工作。在此示例中，窗口恰好是1天，但我可能希望它工作3天，7天，2小时，5秒。

Answer 1

尝试一下：
-首先，按park和date对数据帧进行分组
-通过其唯一值数量汇总to_count

df = pd.DataFrame({
    'park': ['A', 'A', 'A', 'A', 'B', 'B', 'B', 'B'],
    'date': ['2019-01-01', '2019-01-03', '2019-01-05', '2019-01-05',
             '2019-01-01', '2019-01-08', '2019-01-08', '2019-01-10'],
    'to_count': ['Honda', 'Lexus', 'BMW', 'Lexus', 'BMW', 'Lexus', 'Lexus', 'Ford']
})

agg_df = df.groupby(by=['park', 'date']).agg({'to_count': pd.Series.nunique}).reset_index()

Answer 2

我的解决方案不是非常pythonic，但是我认为可以完成工作。

我一次停放一个公园，我将数据帧切成天数偏移（调整天数以获取滚动量），然后将汽车值检索到一个列表中。

使用每天的列表中的汽车，我们可以计算每天的独特汽车总数。

结果是一个列表，您可以根据需要将其转换为数据框。

import pandas as pd
import datetime

# initiate dataframe
df = pd.DataFrame({
    'park': ['A', 'A', 'A', 'A', 'B', 'B', 'B', 'B'],
    'date': ['2019-01-01', '2019-01-03', '2019-01-05', '2019-01-05',
             '2019-01-01', '2019-01-08', '2019-01-08', '2019-01-10'],
    'to_count': ['Honda', 'Lexus', 'BMW', 'Lexus', 'BMW', 'Lexus', 'Lexus', 'Ford']
})
# string to date
df['date'] = pd.to_datetime(df['date'])

result = []
for park in ['A', 'B']:
    # Do one park at a time
    df_park = df[df['park'] == park][['date','to_count']]
    df_park.set_index('date',inplace=True)
    # interate through the dataframe and put results to list.
    for i, v in df_park.iterrows(): 
        # THIS IS YOUR ROLLING VALUE IN DAYS
        days = 1
        # create the starting date
        b = i - datetime.timedelta(days=days)
        # create a list of cars during the period
        li = df_park.loc[b:i].values
        # reduce the list to unique cars
        unique_cars = len(np.unique(li))
        # append the results to the result list
        result.append((park, i.strftime('%B %d, %Y'), unique_cars))

# the final list has duplicates, so use set to drop the dups and re-sort for the result.
sorted(list(set(result)))

结果如下：

[('A', 'January 01, 2019', 1),
('A', 'January 03, 2019', 1),
('A', 'January 05, 2019', 2),
('B', 'January 01, 2019', 1),
('B', 'January 08, 2019', 1),
('B', 'January 10, 2019', 1)]

熊猫可以计算RollingGroupby对象上的字符串类型的列吗？

2 个答案: