[SOLVED] Why does pd.to_datetime('2025175', format='%Y%W%w') and pd.Timestamp.fromisocalendar(2025, 17, 5) gives different output?

Why does pd.to_datetime('2025175', format='%Y%W%w') and pd.Timestamp.fromisocalendar(2025, 17, 5) gives different output?

Why does pd.to_datetime('2025175', format='%Y%W%w') and pd.Timestamp.fromisocalendar(2025, 17, 5) gives different output?

I expected to obtain Timestamp('2025-04-25 00:00:00') for both cases. But the first approach resulted on a Friday one week ahead.

Minimum example

import pandas as pd


friday_datetime = pd.to_datetime('2025175', format='%Y%W%w')
friday_timestamp = pd.Timestamp.fromisocalendar(2025, 17, 5)
assert friday_datetime == friday_timestamp, (friday_datetime, friday_timestamp)

Output:

assert friday_datetime == friday_timestamp, (friday_datetime, friday_timestamp) 
AssertionError: (Timestamp('2025-05-02 00:00:00'), Timestamp('2025-04-25 00:00:00'))

Solution

There are different ways of calculating the week of the year

fromisocalendar uses ISO standard, where week 1 is the week containing January 4th. Or in other words, the first week containing at least 4 days of the new year.
%W uses week containing the first monday of the year as week 1. All days before this are considered to be week 0.

For 2025 these two methods for counting weeks give different results. Ie week 1 in ISO is from 2024-12-30 to 2025-01-05 whereas week 1 following %W logic is from 2025-01-06 to 2025-01-12. Thus, also your week 17 from to_datetime is one week ahead.

If you want pd.to_datetime to behave the same as pd.Timestamp.fromisocalendar use %V as format specifier instead of %W.