Level 1 · Your first DataFrame

What a DataFrame is

A DataFrame is a table: rows of records, named columns, and a label for every row. If you’ve used a spreadsheet, you already know the shape. The difference is that you work with it in code, so every step is written down, repeatable and fast on millions of rows.

Each column of a DataFrame is a Series: a single column of values with one data type, such as all integers or all text. You’ll meet Series constantly, because selecting one column, adding a column or summarising a column all produce one.

On the left of every DataFrame is the index, the row labels. When a table is loaded from a CSV file, the index is simply 0, 1, 2, and so on. It isn’t one of the columns; it’s the row’s address. You’ll see it change when you filter or sort, which matters later.

Looking at the data

You never print a 12,000-row table to read it. You look at a slice:

orders.head()        # first 5 rows
orders.head(10)      # first 10 rows
orders.tail(3)       # last 3 rows
orders.sample(5, random_state=1)   # 5 random rows, the same 5 every time

head shows you how the data starts, which for orders means the oldest records. tail shows the newest. sample is the best way to see “typical” rows, since the start of a file is often special (test orders, early data). random_state fixes the randomness so you and a colleague see the same rows.

These methods return a new DataFrame; they don’t change orders. That’s true of almost every pandas method: you get a result back, and you store it in a variable if you want to keep it.

Reading the orders table

The first rows of orders look like this:

	order_id	customer_id	order_date	status	channel	shipping_fee	coupon_code
0	1001	125	2023-01-01 20:51:19	completed	web	0.00	NaN
1	1002	156	2023-01-02 07:58:46	completed	web	0.00	NaN
2	1003	22	2023-01-02 10:25:27	completed	marketplace	4.99	WELCOME10

A few things to notice straight away:

• Each row is one order, identified by order_id. Knowing what one row represents (the grain of the table) is the most important fact about any table.
• customer_id links each order to a row in customers. You’ll join them in chapter 7.
• coupon_code shows NaN where no coupon was used. NaN (not a number) is how pandas marks a missing value.
• order_date holds a date and a time. It was parsed as a real datetime when the table was loaded, so you can pull out years and months from it. You’ll do that in the cleaning chapter.

Displaying results

In this game, the value in result is what gets checked, and it’s shown as a table under the editor. In a Jupyter notebook, the last expression in a cell is displayed automatically, so you’ll often see analysts write just orders.head() on its own line.

Common mistakes

• Writing orders.head without brackets. That’s the method itself, not its result, and pandas shows something like <bound method NDFrame.head ...>.
• Treating the index as the order ID. Here they happen to move together, but after a filter or a sort they won’t.
• Assuming the first rows are representative. Always look at a sample too.

If you know SQL

orders.head(10) is like SELECT * FROM orders LIMIT 10, with the difference that a DataFrame already has a row order, so the “first ten” are well defined.