Level 5 · Count orders by status

value_counts: how many of each?

For a text or category column, the first question is always “what values are there, and how often?”. value_counts() answers it in one call:

orders["status"].value_counts()

status
completed    10849
cancelled      726
refunded       304
pending        121
Name: count, dtype: int64

The result is a Series: the values become the index and the counts are the data. It’s sorted from most to least frequent, so the top of the list is the answer to “what’s most common?”.

Useful options:

orders["status"].value_counts(normalize=True)   # shares that add up to 1
orders["coupon_code"].value_counts(dropna=False)   # include NaN as its own row
orders["channel"].value_counts(ascending=True)     # least common first

The dropna=False option is important. By default, missing values are silently left out, and for coupon_code the missing rows (no coupon used) are by far the largest group. A share computed without them would be badly wrong.

Related helpers

orders["channel"].unique()     # array(['web', 'marketplace', 'app'], dtype=object)
orders["channel"].nunique()    # 3
orders["customer_id"].nunique()   # how many distinct customers ordered

nunique is how you answer “how many different customers?”, which is not the same as the number of orders.

describe: numbers at a glance

For numeric columns, describe() gives the standard summary:

products["unit_price"].describe()

count     120.00
mean      270.16
std       362.06
min        10.99
25%        51.74
50%       154.00
75%       289.00
max      1659.00

Read it like an analyst. The mean ($270) is far above the median, the 50% line ($154). That gap tells you prices are skewed: a few expensive pieces of furniture pull the average up. For skewed data, the median is usually the better “typical” value to report. The quartiles (25% and 75%) tell you the middle half of products cost between about $52 and $289.

On a whole DataFrame, orders.describe() summarises every numeric column, and orders.describe(include="object") summarises text columns with count, unique, top and freq.

Watch for IDs

describe() will happily report the mean customer_id. It’s meaningless: IDs are labels, not quantities. Ignore summaries of ID columns, or select the columns you actually care about first.

Common mistakes

• Calling value_counts() on the whole DataFrame when you meant one column. orders.value_counts() counts unique combinations of every column.
• Forgetting dropna=False on columns with missing values, then reporting shares that are too high.
• Reporting the mean of a skewed column as “the typical price”.

If you know SQL

orders["status"].value_counts() is SELECT status, COUNT(*) FROM orders GROUP BY status ORDER BY COUNT(*) DESC.