Develop Reference

Python: calculate rolling returns over different frequencies

I have the following DataFrame:
data = [[99330,12,122],[1123,1230,1287],[123,101,812739],[1143,1230123,252],[234,342,4546],[2445,3453,3457],[7897,8657,5675],[46,5675,453],[76,484,3735],[363,93,4568],[385,568,367],[458,846,4847],[574,45747,658468],[57457,46534,4675]]
df1 = pd.DataFrame(data, index=['2022-01-01', '2022-01-02', '2022-01-03', '2022-01-04',
'2022-01-05', '2022-01-06', '2022-01-07', '2022-01-08',
'2022-01-09', '2022-01-10', '2022-01-11', '2022-01-12',
'2022-01-13', '2022-01-14'],
columns=['col_A', 'col_B', 'col_C'])
df1.index = pd.to_datetime(df1.index)
df1:
col_A col_B col_C
2022-01-01 99330 12 122
2022-01-02 1123 1230 1287
2022-01-03 123 101 812739
2022-01-04 1143 1230123 252
2022-01-05 234 342 4546
2022-01-06 2445 3453 3457
2022-01-07 7897 8657 5675
2022-01-08 46 5675 453
2022-01-09 76 484 3735
2022-01-10 363 93 4568
2022-01-11 385 568 367
2022-01-12 458 846 4847
2022-01-13 574 45747 658468
2022-01-14 57457 46534 4675
I am applying the following to get rolling returns:
periodicity_dict = {1:'daily', 7:'weekly'}
df_columns = df1.columns
for key in periodicity_dict:
for col in df_columns:
df1[col+'_rolling']= np.nan
for i in range(key, len(df1[col][df1[col].first_valid_index():df1[col].last_valid_index()])):
df1[col+'_rolling'].iloc[i] = (df1[col].iloc[i] - df[col].iloc[i-key])/df[col].iloc[i-key]
But I am getting the following error: KeyError: 'col_A'.
What am I doing wrong? And is there a better way to do this with less loops?

I think you are looking for something like the shift method (no for-loop is needed):
df1['col_A_rolling'] = (df1['col_A'] - df1['col_A'].shift(7)) / df1['col_A'].shift(7)
OUTPUT:
col_A col_B col_C col_A_rolling
2022-01-01 99330 12 122 NaN
2022-01-02 1123 1230 1287 NaN
2022-01-03 123 101 812739 NaN
2022-01-04 1143 1230123 252 NaN
2022-01-05 234 342 4546 NaN
2022-01-06 2445 3453 3457 NaN
2022-01-07 7897 8657 5675 NaN
2022-01-08 46 5675 453 -0.999537
2022-01-09 76 484 3735 -0.932324
2022-01-10 363 93 4568 1.951220
2022-01-11 385 568 367 -0.663167
2022-01-12 458 846 4847 0.957265
2022-01-13 574 45747 658468 -0.765235
2022-01-14 57457 46534 4675 6.275801

Add Future dates for missing Rows in a Dataframe

How to impute the Missed dates with next Dates in a data frame?
wtg_at1.tail(10)
AmbientTemperatue
Date
818
31.237499
2020-03-28
819
32.865974
2020-03-29
820
32.032558
2020-03-30
821
31.671166
NaN
822
31.389927
NaN
823
31.243660
NaN
824
31.206777
NaN
825
31.241503
NaN
826
31.309531
NaN
827
31.382531
NaN
I am expecting my output data frame something similar to below. After 30th March, I am expecting next dates from 31st March.
AmbientTemperatue
Date
818
31.237499
2020-03-28
819
32.865974
2020-03-29
820
32.032558
2020-03-30
821
31.671166
2020-03-31
822
31.389927
2020-04-01
823
31.243660
2020-04-02
824
31.206777
2020-04-03
825
31.241503
2020-04-04
826
31.309531
2020-04-05
827
31.382531
2020-04-06
I tried below code but not giving desired output.
wtg_at1.append(pd.DataFrame({'Date': pd.date_range(start=wtg_at1.Date.iloc[-8], periods=7, freq='D', closed='right')}))
wtg_at1
AmbientTemperatue
Date
0
32.032558
2017-12-31
1
26.667757
2018-01-01
2
25.655754
2018-01-02
3
25.514013
2018-01-03
4
24.927652
2018-01-04
...
...
...
823
31.243660
NaN
824
31.206777
NaN
825
31.241503
NaN
826
31.309531
NaN
827
31.382531
NaN

If there is only one group of missing values is possible forward filling them and add counter by cumulative sum converted to days timedeltas:
df['Date'] = pd.to_datetime(df['Date'])
df['Date'] = df['Date'].ffill() + pd.to_timedelta(df['Date'].isna().cumsum(), unit='d')
print (df)
AmbientTemperatue Date
818 31.237499 2020-03-28
819 32.865974 2020-03-29
820 32.032558 2020-03-30
821 31.671166 2020-03-31
822 31.389927 2020-04-01
823 31.243660 2020-04-02
824 31.206777 2020-04-03
825 31.241503 2020-04-04
826 31.309531 2020-04-05
827 31.382531 2020-04-06
Another possible idea is reassign values by minimal datetime and length of DataFrame:
df['Date'] = pd.date_range(df['Date'].min(), periods=len(df))
If there is multiple groups with missing values:
print (df)
AmbientTemperatue Date
818 31.237499 2020-03-28
819 32.865974 2020-03-29
820 32.032558 2020-03-30
821 31.671166 NaN
822 31.389927 NaN
823 31.243660 NaN
824 31.206777 2020-05-08
825 31.241503 NaN
826 31.309531 NaN
827 31.382531 NaN
df['Date'] = pd.to_datetime(df['Date'])
m = df['Date'].notna()
s = (~m).groupby(m.cumsum()).cumsum()
df['Date'] = df['Date'].ffill() + pd.to_timedelta(s, unit='d')
print (df)
AmbientTemperatue Date
818 31.237499 2020-03-28
819 32.865974 2020-03-29
820 32.032558 2020-03-30
821 31.671166 2020-03-31
822 31.389927 2020-04-01
823 31.243660 2020-04-02
824 31.206777 2020-05-08
825 31.241503 2020-05-09
826 31.309531 2020-05-10
827 31.382531 2020-05-11

How to get values for the next month for a selected column from a pandas data frame with date time index

I have the below data frame (date time index, with all working days in us calender)
import pandas as pd
from pandas.tseries.holiday import USFederalHolidayCalendar
from pandas.tseries.offsets import CustomBusinessDay
import random
us_bd = CustomBusinessDay(calendar=USFederalHolidayCalendar())
dt_rng = pd.date_range(start='1/1/2018', end='12/31/2018', freq=us_bd)
n1 = [round(random.uniform(20, 35),2) for _ in range(len(dt_rng))]
n2 = [random.randint(100, 200) for _ in range(len(dt_rng))]
df = pd.DataFrame(list(zip(n1,n2)), index=dt_rng, columns=['n1','n2'])
print(df)
n1 n2
2018-01-02 24.78 197
2018-01-03 23.33 176
2018-01-04 33.19 128
2018-01-05 32.49 110
... ... ...
2018-12-26 31.34 173
2018-12-27 29.72 166
2018-12-28 31.07 104
2018-12-31 33.52 184
[251 rows x 2 columns]
For each row in column n1 , how to get values from the same column for the same day of next month? (if value for that exact day is not available (due to weekends or holidays), then should get the value at the next available date. ). I tried using df.n1.shift(21), but its not working as the exact working days at each month differ.
Expected output as below
n1 n2 next_mnth_val
2018-01-02 25.97 184 28.14
2018-01-03 24.94 133 27.65 # three values below are same, because on Feb 2018, the next working day after 2nd is 5th
2018-01-04 23.99 143 27.65
2018-01-05 24.69 182 27.65
2018-01-08 28.43 186 28.45
2018-01-09 31.47 104 23.14
... ... ... ...
2018-12-26 29.06 194 20.45
2018-12-27 29.63 158 20.45
2018-12-28 30.60 148 20.45
2018-12-31 20.45 121 20.45
for December , the next month value should be last value of the data frame ie, value at index 2018-12-31 (20.45).
please help.

This is an interesting problem. I would shift the date by 1 month, then shift it again to the next business day:
df1 = df.copy().reset_index()
df1['new_date'] = df1['index'] + pd.DateOffset(months=1) + pd.offsets.BDay()
df.merge(df1, left_index=True, right_on='new_date')
Output (first 31st days):
n1_x n2_x index n1_y n2_y new_date
0 34.82 180 2018-01-02 29.83 129 2018-02-05
1 34.82 180 2018-01-03 24.28 166 2018-02-05
2 34.82 180 2018-01-04 27.88 110 2018-02-05
3 24.89 186 2018-01-05 25.34 111 2018-02-06
4 31.66 137 2018-01-08 26.28 138 2018-02-09
5 25.30 162 2018-01-09 32.71 139 2018-02-12
6 25.30 162 2018-01-10 34.39 159 2018-02-12
7 25.30 162 2018-01-11 20.89 132 2018-02-12
8 23.44 196 2018-01-12 29.27 167 2018-02-13
12 25.40 153 2018-01-19 28.52 185 2018-02-20
13 31.38 126 2018-01-22 23.49 141 2018-02-23
14 30.90 133 2018-01-23 25.56 145 2018-02-26
15 30.90 133 2018-01-24 23.06 155 2018-02-26
16 30.90 133 2018-01-25 24.95 174 2018-02-26
17 29.39 138 2018-01-26 21.28 157 2018-02-27
18 32.94 173 2018-01-29 20.26 189 2018-03-01
19 32.94 173 2018-01-30 22.41 196 2018-03-01
20 32.94 173 2018-01-31 27.32 149 2018-03-01
21 28.09 119 2018-02-01 31.39 192 2018-03-02
22 32.21 199 2018-02-02 28.22 151 2018-03-05
23 21.78 120 2018-02-05 34.82 180 2018-03-06
24 28.25 127 2018-02-06 24.89 186 2018-03-07
25 22.06 189 2018-02-07 32.85 125 2018-03-08
26 33.78 121 2018-02-08 30.12 102 2018-03-09
27 30.79 137 2018-02-09 31.66 137 2018-03-12
28 29.88 131 2018-02-12 25.30 162 2018-03-13
29 20.02 143 2018-02-13 23.44 196 2018-03-14
30 20.28 188 2018-02-14 20.04 102 2018-03-15

How to join DataFrame with multiple conditions on different columns?

I have two data-frames as follows:
mydata1:
ID X1 X2 Date1
002 324 634 2016-01-01
002 334 534 2016-01-14
002 354 834 2016-01-30
004 543 843 2017-02-01
004 923 043 2017-04-15
005 032 212 2015-09-01
005 523 843 2017-09-15
005 212 222 2015-10-1
mydata2:
ID Y1 Y2 Date2
002 1224 234 2016-01-04
002 1254 249 2016-01-28
004 321 212 2016-12-01
005 1121 222 2017-09-13
I want to merge these two data-frames based on ID and the Date where the difference between Date1 --dataframe1-- and Date2 --indataframe2--is less than 15. So, my desired data-frame as an output should be like this:
ID X1 X2 Date1. Y1. Y2. Date2
002 324 634 2016-01-01. nan. nan. nan
002 334 534 2016-01-14 1224 234 2016-01-04
002 354 834 2016-01-30. 1254 249 2016-01-28
004 543 843 2017-02-01 321 212 2015-12-01
004 923 043 2017-04-15. nan nan. nan
005 032 212 2015-09-01 nan nan. nan
005 523 843 2015-09-15. 1121 222 2017-09-13
005 212 222 2015-10-1. nan nan. nan

So your desired output is slightly wrong since one of the values is 2 years older than the joined value.
First we perform a join:
f = df.merge(df1, how='left', on='ID')
ID X1 X2 Date1 Y1 Y2 Date2
0 2 324 634 2016-01-01 1224 234 2016-01-04
1 2 334 534 2016-01-14 1224 234 2016-01-04
2 2 354 834 2016-01-30 1224 234 2016-01-04
3 4 543 843 2017-02-01 321 212 2016-12-01
4 4 923 43 2017-04-15 321 212 2016-12-01
5 5 32 212 2015-09-01 1121 222 2015-09-13
6 5 523 843 2015-09-15 1121 222 2015-09-13
7 5 212 222 2015-10-1 1121 222 2015-09-13
Then we create a boolean mask:
mask = (pd.to_datetime(f['Date1'], format='%Y-%m-%d') - pd.to_datetime(f['Date2'], format='%Y-%m-%d')).apply(lambda i: i.days <= 15 and i.days > 0)
0 False
1 True
2 False
3 False
4 False
5 False
6 True
7 False
Then we set it to nan where the condition does not match:
f.loc[~mask, ['Y1', 'Y2', 'Date2']] = np.nan
ID X1 X2 Date1 Y1 Y2 Date2
0 2 324 634 2016-01-01 NaN NaN NaN
1 2 334 534 2016-01-14 1224.0 234.0 2016-01-04
2 2 354 834 2016-01-30 NaN NaN NaN
3 4 543 843 2017-02-01 NaN NaN NaN
4 4 923 43 2017-04-15 NaN NaN NaN
5 5 32 212 2015-09-01 NaN NaN NaN
6 5 523 843 2015-09-15 1121.0 222.0 2015-09-13
7 5 212 222 2015-10-1 NaN NaN NaN

how to subset pandas dataframe on date

I have a pandas DataFrame like this..
order_id buyer_id item_id time
537 79 93 2016-01-04 10:20:00
540 191 93 2016-01-04 10:30:00
556 251 82 2016-01-04 13:39:00
589 191 104 2016-01-05 10:59:00
596 251 99 2016-01-05 13:48:00
609 79 106 2016-01-06 10:39:00
611 261 97 2016-01-06 10:50:00
680 64 135 2016-01-11 11:58:00
681 261 133 2016-01-11 12:03:00
682 309 135 2016-01-11 12:08:00
I want to get all the buyer_ids present before 6th jan 2016 but not after 6th Jan 2016
so, it should return me buyer_id 79
I am doing following in Python.
df.buyer_id[(df['time'] < '2016-01-06')]
This returns me all the buyer ids before 6th jan 2016 but how to check for the condition if its not present after 6th jan ? Please help

IIUC you could use isin method to achieve what you want:
df.time = pd.to_datetime(df.time)
In [52]: df
Out[52]:
order_id buyer_id item_id time
0 537 79 93 2016-01-04 10:20:00
1 540 191 93 2016-01-04 10:30:00
2 556 251 82 2016-01-04 13:39:00
3 589 191 104 2016-01-05 10:59:00
4 596 251 99 2016-01-05 13:48:00
5 609 79 106 2016-01-06 10:39:00
6 611 261 97 2016-01-06 10:50:00
7 680 64 135 2016-01-11 11:58:00
8 681 261 133 2016-01-11 12:03:00
9 682 309 135 2016-01-11 12:08:00
exclude = df.buyer_id[(df['time'] > '2016-01-06')]
select = df.buyer_id[(df['time'] < '2016-01-06')]
In [53]: select
Out[53]:
0 79
1 191
2 251
3 191
4 251
Name: buyer_id, dtype: int64
In [54]: exclude
Out[54]:
5 79
6 261
7 64
8 261
9 309
Name: buyer_id, dtype: int64
In [55]: select[~select.isin(exclude)]
Out[55]:
1 191
2 251
3 191
4 251
Name: buyer_id, dtype: int64

You could use:
df.groupby('buyer_id').apply(lambda x: True if (x.time < '01-06-2016').any() and not (x.time > '01-06-2016').any() else False)
buyer_id
64 False
79 False
191 True
251 True
261 False
309 False
dtype: bool