Online Numpy Compiler

NumPy is the array library that most numerical Python code is built on. Its ndarray holds numbers of one type in a grid with any number of dimensions, and math on a whole array runs in compiled code, which is much faster than a loop over a Python list. pandas, SciPy, scikit-learn and matplotlib all build on it, and people use it directly for data analysis, simulations and linear algebra. This page is an online NumPy compiler: the code runs in your browser, so you can try it without installing anything. Run the example first, then paste any snippet below into a new cell to try it.

What the example does

np.array() turns a Python list into an array, here two arrays of six integers. np.add() and np.multiply() work element by element, pairing the first value of one array with the first of the other, and so on. They give the same result as array1 + array2 and array1 * array2. np.dot() adds up those products: 6 + 10 + 12 + 12 + 10 + 6 = 56. np.reshape(array1, (2, -1)) arranges the six values in two rows, and -1 tells NumPy to work out the number of columns. np.mean() and np.max() reduce an array to a single number.

Create arrays

These functions build arrays of a given size. Every array has a shape, a number of dimensions (ndim) and one dtype shared by all its values:

import numpy as np

print(np.arange(0, 10, 2))       # like range(): start, stop, step
print(np.linspace(0, 1, 5))      # 5 evenly spaced points, end included
print(np.zeros((2, 3)))          # 2 rows, 3 columns of 0.0

grid = np.arange(12).reshape(3, 4)
print(grid)
print(grid.shape, grid.ndim, grid.dtype)

Filter values with a condition

Comparing an array with a value gives an array of True and False, called a mask. Indexing with the mask keeps only the matching values, and assigning through it changes them in place:

import numpy as np

temps = np.array([18.5, 22.1, 25.3, 19.8, 30.2, 27.6, 16.4])

hot = temps > 25
print(hot)                       # one True/False per value
print(temps[hot])                # only the values where the mask is True
print(hot.sum(), "hot days")     # True counts as 1
print(np.where(temps > 25, "hot", "mild"))

temps[temps < 18] = 18           # set every value below 18 to 18
print(temps)

Work along rows and columns

Functions that reduce an array take an axis: axis=0 works down the columns and axis=1 across the rows. The data comes from np.random.default_rng(), and a fixed seed gives the same numbers on every run:

import numpy as np

rng = np.random.default_rng(seed=42)
scores = rng.integers(50, 101, size=(4, 3))   # 4 students, 3 tests
print(scores)

print(scores.mean(axis=0))   # average of each column (per test)
print(scores.max(axis=1))    # best score of each row (per student)

# Broadcasting: subtract each test's average from its column
print(scores - scores.mean(axis=0))

integers() leaves out the upper bound, so scores run from 50 to 100. The last line uses broadcasting: NumPy repeats the three column averages for every row, so no loop is needed.

Multiply matrices and solve equations

On two arrays, * multiplies element by element. For a matrix product, use @. The np.linalg module has the usual linear algebra routines:

import numpy as np

A = np.array([[1, 2], [3, 4]])
B = np.array([[5, 6], [7, 8]])

print(A * B)    # element by element
print(A @ B)    # matrix product, same as np.matmul(A, B)
print(A.T)      # transpose
print(np.linalg.inv(A))

# Solve 2x + 3y = 5 and 4x - y = 3
coefficients = np.array([[2, 3], [4, -1]])
print(np.linalg.solve(coefficients, np.array([5, 3])))

The SciPy example solves the same two equations with scipy.linalg.solve.

Good to know

• In this browser build, NumPy's default integer type is 32-bit (int32), because Python runs as 32-bit WebAssembly. A 64-bit desktop install uses int64. Integer results past about 2.1 billion wrap around without an error: np.arange(100000).sum() gives 704982704 instead of 4999950000. Pass dtype=np.int64 when numbers get large.
• A slice such as a[1:3] is a view of the original array, so changing the slice changes a too. Call .copy() when you need an independent array. Indexing with a mask always returns a copy.