Counting categories with value_counts

value_counts() is a Series method that returns how many rows carry each distinct value, as a new Series whose index is the values and whose numbers are the counts, sorted from the most common down.

Last updated: 12 Oct, 2026 · pandas 3.0 · seaborn 0.13 · matplotlib 3.11

One chart, one question, one observation drew its chart from one call to value_counts(). It is the first thing to run on any text or coded column, and the answer it gives has more in it than the counts.

Syntax: Series.value_counts() counts the distinct values. normalize=True turns counts into shares, dropna=False counts the missing values as a group of their own, and sort=False leaves the values in the order pandas found them.

counts = df['Country'].value_counts()   # a Series: value -> count
counts.index                            # the distinct values
counts.values                           # the counts, as a numpy array
df['Country'].value_counts(normalize=True)   # shares that add up to 1
df['Country'].value_counts(dropna=False)     # NaN counted as a group

Loading the Zomato data

Every example here starts from the merged Zomato frame. The zip recipe is the one Reading a CSV that is not UTF-8 built, and the join on the country code is the one Merging a lookup table built. Run this block first, then the example under it.

import io, zipfile, urllib.request
import pandas as pd
import matplotlib
matplotlib.use('svg')
import matplotlib.pyplot as plt

url = ("https://raw.githubusercontent.com/krishnaik06/"
       "5-Days-Live-EDA-and-Feature-Engineering/main/Zomatodataset.zip")
zf = zipfile.ZipFile(io.BytesIO(urllib.request.urlopen(url).read()))
with zf.open('zomato.csv') as fh:
    df = pd.read_csv(fh, encoding='latin-1')
with zf.open('Country-Code.xlsx') as fh:
    codes = pd.read_excel(fh)
df = pd.merge(df, codes, on='Country Code', how='left')

Reading the three parts of the answer

The returned object is a Series, which means it has an index and values like any other. The index holds the country names and the values hold the counts, and both halves are useful on their own: the index becomes the labels of a chart, the values become its bars.

counts = df['Country'].value_counts()
print("type:", type(counts).__name__)
print(counts.head(3).to_string())
print("labels:", counts.index[:3].tolist())
print("numbers:", counts.values[:3].tolist())
print("as a share:")
print(df['Country'].value_counts(normalize=True).head(3).round(4).to_string())

type: Series
Country
India             8652
United States      434
United Kingdom      80
labels: ['India', 'United States', 'United Kingdom']
numbers: [8652, 434, 80]
as a share:
Country
India             0.9059
United States     0.0454
United Kingdom    0.0084

What the counts of Country say

• India holds 8,652 of the 9,551 rows, a share of 0.9059. The dataset is an Indian dataset with a thin international tail.
• The second country is 20 times smaller. United States has 434 rows, United Kingdom 80. Any average computed over the whole table is an Indian average.
• The index and the values come apart cleanly. counts.index is the list of country names and counts.values is the matching array of numbers, which is what a pie or bar chart needs.

Counting the missing values as a group

By default value_counts() leaves NaN out, so the counts add up to the number of non-missing rows rather than to len(df). dropna=False adds one more group for the missing values. On a column with few nulls the difference is one line in the output and a few rows in the total, which is enough to make a share wrong.

print("groups counted by default:", len(df['Cuisines'].value_counts()))
print("groups counted with dropna=False:",
      len(df['Cuisines'].value_counts(dropna=False)))
print("rows with no cuisine listed:", int(df['Cuisines'].isnull().sum()))
print("counts add up to:", int(df['Cuisines'].value_counts().sum()),
      "out of", len(df))

groups counted by default: 1825
groups counted with dropna=False: 1826
rows with no cuisine listed: 9
counts add up to: 9542 out of 9551

1,825 distinct cuisine listings become 1,826 groups once the missing ones are counted, and the default counts add up to 9,542 rather than 9,551. The nine rows with no cuisine are the ones The missing value map found.

Charting the ten most common values

head(10) on the result keeps the ten biggest groups, and plot.barh() draws them as horizontal bars, which is the readable choice when the labels are long. sort_values() before plotting puts the biggest bar at the top, because a horizontal bar chart draws its first row at the bottom.

top = df['Cuisines'].value_counts().head(10)
print(top.to_string())

fig, ax = plt.subplots(figsize=(7, 4.5))
top.sort_values().plot.barh(ax=ax, color='#2e9e5b')
ax.set_xlabel('restaurants')
ax.set_title('Ten most common cuisine listings')
fig.savefig('chart.svg')

Cuisines
North Indian                      936
North Indian, Chinese             511
Chinese                           354
Fast Food                         354
North Indian, Mughlai             334
Cafe                              299
Bakery                            218
North Indian, Mughlai, Chinese    197
Bakery, Desserts                  170
Street Food                       149

What the cuisine counts say

• North Indian alone is 936 restaurants, and it appears again inside three of the other nine listings. The column holds a comma separated list, not one category.
• 1,825 distinct listings over 9,551 rows means the top ten cover only a fifth of the table. A column like this needs splitting on the comma before it can be counted properly.
• Two listings tie at 354. value_counts() keeps ties in the order it found them, so do not read the order of equal counts as a ranking.

The call that was removed

Older notebooks count through the top level function pd.value_counts(series) instead of the method. It was removed in pandas 2.2, and the attribute is gone, so the line fails at once rather than returning something wrong.

import pandas as pd

s = pd.Series(['India', 'India', 'UAE'])
pd.value_counts(s)

Traceback (most recent call last):
  File "main.py", line 4, in <module>
    pd.value_counts(s)
AttributeError: module 'pandas' has no attribute 'value_counts'

The fix is to call it on the Series: s.value_counts(). Every argument is the same.

value_counts vs groupby size

Where value_counts is the first call

• Checking a column's cardinality. len(df[c].value_counts()) says how many distinct values a column has, which decides how it can be encoded.
• Finding the mode. The first index entry is the most frequent value, which is what Frequent category imputation fills with.
• Spotting a dirty level. Two spellings of one category show up as two rows with small counts, which no summary statistic would reveal.

Related

• Previous: One chart, one question, one observation
• Next: Pie and bar charts with a top-N trim
• See also: Frequent category imputation
• Reference: Series.value_counts in the pandas docs