NDArray

An ndarray is an N-dimensional array in which all elements share a common data type. It is the core data structure used by NumPy for storing and manipulating large numerical datasets. An ndarray encapsulates both the data buffer and metadata describing its shape, datatype, and memory layout.

Key attributes include shape (a tuple describing lengths along each axis), ndim (the number of axes), size (total number of elements), dtype (the data type of elements), itemsize (bytes per element), and nbytes (total bytes consumed). Strides describe how many bytes to skip to move to the next element along each axis; they determine whether the array is contiguous in memory and whether it is viewed in C (row-major) or Fortran (column-major) order. Flags indicate properties such as C_CONTIGUOUS or F_CONTIGUOUS.

Elements are stored contiguously by default, though operations such as slicing or transpose can create views—arrays that share the same data buffer but present a different shape or order. Copying may be required to obtain an independent, contiguous copy.

Ndarrays are created from Python sequences with array or asarray, or via factory helpers such as zeros, ones, empty, full, arange, and linspace. Common operations include indexing and slicing, reshaping, transposing, flattening, and broadcasting. Arithmetic and mathematical functions are implemented as vectorized operations or ufuncs, applying elementwise operations efficiently at C speed.

Broadcasting allows arithmetic between arrays of different shapes under compatibility rules, enabling operations without explicit replication of data.

Ndarrays provide memory efficiency and speed for numerical workloads but require homogeneous element types and strides compatibility. They form the foundation for higher-level structures in scientific computing ecosystems and are widely used in data analysis and machine learning pipelines.