Learn NumPy

NumPy's core object: a fixed-size, typed, n-dimensional grid of numbers stored in one block of memory.

Build arrays without typing values: zeros, ones, full, arange, linspace and the *_like helpers.

Every array has one dtype that fixes how many bytes each element uses and how they are interpreted.

Change how the same data is arranged into axes with reshape, ravel, transpose and new axes.

Arrays whose elements are records with named, typed fields, matching C structs and binary formats.

Pick elements, rows, columns and sub-blocks with integer indexes and start:stop:step slices per axis.

Select, count and modify elements that meet a condition using boolean arrays as indexes.

Index with arrays of positions to gather, reorder or scatter arbitrary elements.

Replace Python loops with whole-array expressions that run in compiled code.

np.sqrt, np.exp, np.maximum and friends: fast element-wise functions with extra powers like out= and where=.

Reduce arrays with sum, mean, min, max, std, argmax, and choose which axis to collapse.

The rules that let arrays of different shapes combine element-wise without copying data.

Express dot products, transposes, traces and batched contractions in one subscript string.

Know which operations share memory with the original array and which make independent copies.

How shape, strides and C/Fortran order map n-dimensional indexes onto one flat buffer.

Turn loop-heavy code into array code: masks, cumsum/diff, bincount, reduceat, and avoiding temporaries.

Use np.random.default_rng for reproducible, high-quality random numbers and sampling.

Matrix products with @, solving systems, inverses, norms, decompositions and batched matrices.

sort, argsort, searchsorted, partition and unique: ordering data and finding things quickly.

Join arrays with concatenate and stack, and cut them apart with split and array_split.

Persist arrays with .npy/.npz, read text files with loadtxt/genfromtxt, and memory-map big files.

Rounding error, machine epsilon, comparing floats, cancellation and summation accuracy in NumPy.

Represent and compute around missing values with NaN, nan-aware functions and numpy.ma.

What an ndarray is in C (data pointer, dims, strides, descr, base, flags) and how it shares memory with other code.

Type resolution, inner loops, the iterator, generalized ufuncs, and ways to build your own.

How other array libraries plug into NumPy (__array__, __array_ufunc__, __array_function__) and how to write library-agnostic code.

Where NumPy's speed comes from: CPU-dispatched SIMD loops, multithreaded BLAS, the GIL and free-threaded Python.