How do I count the occurrence of a certain item in an ndarray?

Beyond the basics: alternative counting methods

The initial np.sum method is the fastest route, but a few other techniques are versatile for specific scenarios. Let's explore them:

Count occurrences and get unique items with numpy.unique

Generate an inventory of unique items with corresponding counts:

unique, counts = np.unique(arr, return_counts=True)
occurrences = dict(zip(unique, counts))


numpy.unique gives both unique items and their counts, a double boon when you need distinct elements and frequency. A Python dict eases the item-quantity retrieval.

Count only non-negative integers using numpy.bincount

An array with non-negative integers is best dealt with using:

counts = np.bincount(arr)


Where np.bincount serves as a efficient tool for counting the frequency of each non-negative integer.

Use a non-NumPy method with collections.Counter

A worthwhile Python standard library alternative is collections.Counter:

from collections import Counter
counter = Counter(arr.ravel())  # you shall not pass without a counter


The Counter method formulates a dictionary-like object for each unique item in the array, providing an easy way to keep tally.

Visualization

Let's explore this concept with a hypothetical ndarray.

import numpy as np

# Array representation of your items
items = np.array(['🌹', '🌻', '🌹', '🌼', '🌹', '🌻', '🌼'])

# Counting the roses
roses_count = np.sum(items == '🌹')  # stop and count the roses


Your counting result:

Item Type	Occurrence
🌹	3

Tactical counting techniques

There are cases where you need to count items satisfying specific conditions. Here's how to leverage NumPy's capabilities:

Count occurrences with multiple conditions

Count items that fit multiple criteria using:

count = np.sum((arr > x) & (arr < y))  # Goldilocks counting: not too big, not too small


This code returns the count of elements in arr that are both greater than x and less than y.

Efficiently count non-zero elements with numpy.count_nonzero

If your objective is to count only ones or non-zero elements:

count = np.count_nonzero(arr)  # zero is not a hero today


This direct and optimized method utilizes np.count_nonzero.

Powering through with NumPy

You can get more from NumPy, with advanced techniques to enhance computation and provide robust results:

Update array elements during counting

Sometimes, you may want to update array elements based on their frequency:

unique, counts = np.unique(arr, return_counts=True)
arr[np.isin(arr, unique[counts < threshold])] = new_value  # threshold: the 'You shall not pass!' of numbers


The np.isin function checks for elements below a certain threshold, replacing them with a new_value.

Count occurrences using NumPy in tandem with Pandas

An interplay between NumPy and Pandas can offer a useful perspective:

import pandas as pd

# Turning the array into a Pandas series
series = pd.Series(arr.ravel())  # Ravel: not just for classical composers

# Using Pandas to count occurrences
counts = series.value_counts()  # who's counting? Pandas!


This shows that Pandas.value_counts is handy when dealing with tabular data and offers an alternative approach. External tools like Pandas complement NumPy's powerful computational abilities.