[PYTHON] [Note] pandas unstack

TL;DR

ʻUnstack` was useful for making the multiIndex series easier to see.

group by post-processing

DataFrame of pandas will have index MultiIndex if you do group by in multiple columns. I'm a little clogged up to process, so I'll write what I did as a memorandum.

environment

here,

• Python == 3.8
• pandas == 1.1.3

I'm running on.

Data preparation

For example, if you have the following data:

import datetime
import random
import pandas as pd

item_list = ['A', 'A', 'A', 'B', 'C','C', 'D']
data_records = []
ts = datetime.datetime.now()
for _ in range(1000):
    ts += datetime.timedelta(seconds=random.randint(200, 3600))
    data_records.append({
        'ts': ts,
        'wday': ts.weekday(),
        'item': random.choice(item_list),
        'qty': random.randint(1, 5)
    })
df = pd.DataFrame(data_records)

As df You should get something like this.

here,

• ts: timestamp
• wday: day of the week
• item: Product (ID)
• qty: Quantity

Imagine something like a log of an EC site.

Thing you want to do

Now suppose you want to see how many items sell in total for each day of the week. Actually, it is normal to specify the period with ts, but aside from that, I think that you will do the following.

df.groupby(['wday', 'item']).qty.sum()

Then you will get something like this: It's not bad, but it's also hard to see. Here, if you do ʻunstack`,

df.groupby(['wday', 'item']).qty.sum().unstack()

have become.

reference

For more information, see Pandas Official Documentation.