A First Look: head, info and describe
head, tail and sample show a few of a DataFrame's rows, info lists every column's type and how many values are there (the first hint of missing values), and describe sums up each number column in eight numbers, from count and mean to the quartiles, with include='all' adding the text columns.
Apprentice · Python · Updated
After this lesson you can
- Look at a DataFrame's rows with head, tail and sample, and read info's dtypes and non-null counts to spot missing values
- Read describe's summary of a number column in plain words, count, mean, std, min, the quartiles and max
Key takeaways
- head() shows the first five rows, tail() the last, and sample() a few chosen at random, so you see real values before you trust a column.
- info() lists every column with its dtype and its non-null count. A count lower than the number of rows means values are missing.
- describe() gives count, mean (the average), std (how far values usually sit from the average), min, the quartiles and max for each number column.
- The 25% row is the value a quarter of the values are at or below, 50% is the median, the middle value, and 75% is the value three quarters are at or below.
- describe(include="all") adds the text columns, with how many different values each has, the most common one and how often it appears.
A new dataset lands on your desk: a class project's survey, a month of a shop's sales, a sensor's readings. Before you average or chart anything, you spend a few minutes getting to know it: what the rows look like, what type each column is, whether anything is missing, and what range the numbers cover. Three methods do most of that work, and analysts run them on every table they open.
The example is Sunnyside Farmers Market's sales for the 2026 season, sales.csv: one row for each product a vendor sold on a Saturday, 1,900 rows, with the columns date, vendor_id, product, category, units and price. The market is our own made-up data.
Peek at the rows: head, tail and sample
Load the file with pd.read_csv (pandas is imported as pd, read "p d") and look at the first rows with head(), five unless you ask for more:
import pandas as pd
sales = pd.read_csv("sales.csv")
sales.head()
date vendor_id product category units price
0 2026-04-04 V01 lettuce produce 21.0 3.0
1 2026-04-04 V01 kale produce 18.0 3.5
2 2026-04-04 V01 spinach produce 21.0 4.0
3 2026-04-04 V02 honey jar honey 20.0 12.0
4 2026-04-04 V02 honeycomb honey 4.0 15.0
tail() shows the last rows. The first and last rows are often the oldest and newest, so sample() picks rows at random from anywhere in the table; random_state=1 makes it pick the same rows every time, so your notebook gives the same answer when it runs again:
sales.sample(3, random_state=1)
date vendor_id product category units price
1857 2026-10-31 V05 birdhouse crafts 3.0 30.0
414 2026-05-30 V03 cinnamon roll baked goods 59.0 4.0
1162 2026-08-15 V12 salad greens produce 34.0 4.0
info: types and gaps
info() describes every column at once:
sales.info()
<class 'pandas.DataFrame'>
RangeIndex: 1900 entries, 0 to 1899
Data columns (total 6 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 date 1900 non-null str
1 vendor_id 1900 non-null str
2 product 1900 non-null str
3 category 1900 non-null str
4 units 1888 non-null float64
5 price 1900 non-null float64
dtypes: float64(2), str(4)
memory usage: 89.2 KB
Read it line by line: 1,900 rows; each column's name, its Non-Null Count (how many values are there; "null" means missing), and its dtype. Every column has all 1,900 values except units, with 1,888: twelve vendors never reported how many they sold. That's also why units is float64, numbers with decimals, though they're all whole: in pandas, a number column with gaps is stored as decimals. (In pandas 3, text shows as str; older tutorials show object.)
count() gives the same non-null counts as numbers, so the rows minus the count is how many are missing in each column:
len(sales) - sales.count()
date 0
vendor_id 0
product 0
category 0
units 12
price 0
dtype: int64
describe: the numbers in plain words
describe() sums up each number column. round(2) keeps it to two decimals:
sales.describe().round(2)
units price
count 1888.00 1900.00
mean 22.42 10.63
std 19.46 10.65
min 1.00 0.50
25% 8.00 3.25
50% 17.00 7.00
75% 31.00 12.00
max 120.00 45.00
Here's each row for units, in plain words:
- count: 1,888 values are there (the 12 gaps are left out).
- mean: the average: a vendor sold about 22 of a product on a Saturday.
- std: the standard deviation, how far values usually sit from the average. About 19 here: most rows sold within about 19 units of 22, between 3 and 41.
- min and max: the smallest and largest, 1 and 120.
- 25%: read "25 percent of the values are at or below this": a quarter of the rows sold 8 units or fewer. These cut points are called quartiles.
- 50%: half sold 17 or fewer. That's the median, the middle value.
- 75%: three quarters sold 31 or fewer.
Notice the mean, 22, is well above the median, 17. A few cheap, popular products, such as sweet corn at 75 cents an ear, sell around a hundred on a Saturday and pull the average up; the median stays in the middle. When the two disagree, a few large values are worth a look.
The text columns too: include="all"
describe() skips text columns unless you ask for them:
sales.describe(include="all")
date vendor_id product category units price
count 1900 1900 1900 1900 1888.000000 1900.000000
unique 31 15 80 5 NaN NaN
top 2026-09-19 V03 honey jar produce NaN NaN
freq 73 180 60 654 NaN NaN
mean NaN NaN NaN NaN 22.416843 10.625395
std NaN NaN NaN NaN 19.463348 10.645001
min NaN NaN NaN NaN 1.000000 0.500000
25% NaN NaN NaN NaN 8.000000 3.250000
50% NaN NaN NaN NaN 17.000000 7.000000
75% NaN NaN NaN NaN 31.000000 12.000000
max NaN NaN NaN NaN 120.000000 45.000000
For text, unique is how many different values there are, top the most common one, and freq how often it appears: 31 Saturdays, 15 vendors, 80 products, and 654 of the rows are produce. NaN, read "not a number", fills the cells where a summary doesn't apply, such as the average of a vendor's id.
Check yourself
2 questions to try on the lesson page.
So what
head, info and describe take a minute and tell you what's in a table, what's missing and what's normal, before any number you report depends on it.