Transforming every n rows of pandas Series into n columns of a DataFrame

user_9857631 0
Viewed 106

I'm trying to transform a pandas Series like:

Date Value
2020-01-01 -1175
2020-01-02 -475
2020-01-03 1945
2020-01-06 -1295
2020-01-07 -835
2020-01-08 -785
2020-01-09 895
2020-01-10 -665

into a pandas DataFrame like:

date 0 1 2 3 4
2020-01-01 -1175 -475 1945 -1295 -665
2020-01-02 -475 1945 -1295 -835 -785
2020-01-03 1945 -1295 -835 -785 895
2020-01-06 -1295 -835 -785 895 -665

Every 5 (or n) rows of the Series forms one row in the DataFrame.

Sample data along with my current (ugly but working) code is as follows:

import pandas as pd
srs = pd.Series(index=pd.DatetimeIndex(pd.date_range(start="2020-01-01",end="2020-1-10",freq="B")), 
                data=[-1175,-475,1945,-1295,-835,-785,895,-665])
n = 5
df = pd.concat({i: srs.shift(-i) for i in range(n)}, axis=1).dropna()
df = df[range(n)]
df.index = df.index.droplevel(level=0)

I was wondering if there is a better/neater/nicer way to do this?

pandas python
1 Answers

Try with sliding_window_view from numpy:

n = 5
v = np.lib.stride_tricks.sliding_window_view(srs.values, n)
df = pd.DataFrame(v, index=srs.index[:v.shape[0]])

df:

               0     1     2     3    4
2020-01-01 -1175  -475  1945 -1295 -835
2020-01-02  -475  1945 -1295  -835 -785
2020-01-03  1945 -1295  -835  -785  895
2020-01-06 -1295  -835  -785   895 -665

Complete Working Example:

import numpy as np
import pandas as pd

srs = pd.Series(index=pd.DatetimeIndex(
    pd.date_range(start="2020-01-01", end="2020-1-10", freq="B")
), data=[-1175, -475, 1945, -1295, -835, -785, 895, -665])

n = 5
v = np.lib.stride_tricks.sliding_window_view(srs.values, n)
df = pd.DataFrame(v, index=srs.index[:v.shape[0]])

print(df)
user_154978880
Related