Level 4 · Which weekday sells most?

Dates are not text

A date stored as text, '2025-03-14 19:02:11', can’t tell you its weekday, can’t be compared safely, and sorts correctly only by luck of the format. Convert it to a real datetime:

pd.to_datetime(orders["order_date"])

In this game the date columns were parsed when the tables were loaded, so order_date already has dtype datetime64[ns], and pd.to_datetime simply returns it unchanged. On the job, a plain pd.read_csv gives you text, so converting is usually the first line of any date work. You can do it at load time with pd.read_csv(path, parse_dates=["order_date"]).

When some values are malformed, errors="coerce" turns them into NaT instead of failing, and a format= string makes parsing strict and fast:

pd.to_datetime(raw_dates, format="%Y-%m-%d", errors="coerce")

Then count the NaT values to see how many dates didn’t parse. Coercing silently and moving on hides data problems.

The .dt accessor

Datetime columns have a .dt accessor, like .str for text:

d = orders["order_date"]
d.dt.year             # 2025
d.dt.month            # 3
d.dt.day_name()       # 'Friday'
d.dt.dayofweek        # 0 = Monday ... 6 = Sunday
d.dt.hour             # 19
d.dt.date             # the date part only
d.dt.to_period("M")   # 2025-03, a monthly period
d.dt.strftime("%Y-%m")   # '2025-03' as text

Filtering by date works with comparisons and between, and pandas accepts date strings on the other side:

orders[orders["order_date"] >= "2025-01-01"]
orders[orders["order_date"].dt.year == 2025]

Sorting days in calendar order

Day names sort alphabetically (Friday, Monday, Saturday…), which nobody wants on a chart. Group by the day number as well and sort by it:

(
    in_2025.assign(day_num=d.dt.dayofweek, weekday=d.dt.day_name())
    .groupby(["day_num", "weekday"], as_index=False)
    .agg(n_orders=("order_id", "count"))
    .sort_values("day_num")[["weekday", "n_orders"]]
)

An alternative is an ordered categorical: pd.Categorical(names, categories=["Monday", ..., "Sunday"], ordered=True), which also sorts correctly in charts.

Other type fixes

Dates are the most common type problem, but not the only one:

pd.to_numeric(raw_prices, errors="coerce")          # text to numbers, bad values to NaN
web_sessions["customer_id"].astype("Int64")         # float IDs back to integers, keeping missing
products["is_active"].astype(int)                   # True/False to 1/0

Int64 (capital I) is pandas’ nullable integer type. Plain astype(int) fails on a column with missing values, which is why web_sessions.customer_id loads as float in the first place.

Reading the result

Weekend days come out ahead for Harbor & Pine, which fits a home-goods store: people browse furniture and decor when they have time. Before acting on it, check whether the pattern holds in 2023 and 2024 too, and whether it’s driven by a few big sale weekends like Black Friday. One year of weekday counts is a hint, not a law.

Common mistakes

• Sorting by the day name and getting Friday first.
• Comparing dates stored as text with dates as datetimes.
• Using errors="coerce" and never counting how many values became NaT.