ndim

ndim is a term used to describe the number of dimensions, or axes, of an array or tensor. It indicates the dimensionality of the data and influences how the object can be indexed, reshaped, and broadcasted in computations.

In NumPy, ndim is exposed in two closely related ways. The attribute ndarray.ndim on an array returns the number of dimensions. The function numpy.ndim also returns the number of dimensions for an array-like object, including Python scalars or lists converted to arrays. Both approaches yield the same value, which equals the length of the array’s shape.

Examples help illustrate ndim. A scalar produced by numpy.array(5) has shape () and ndim 0. A 1-D vector such as numpy.array([1, 2, 3]) has shape (3,) and ndim 1. A 2-D matrix with shape (4, 5) has ndim 2. A 3-D tensor with shape (2, 3, 4) has ndim 3. In general, ndim equals len(shape), where shape is the tuple of extents along each axis. Axes are indexed from 0 to ndim-1.

ndim can change through operations that add or remove axes. For example, numpy.expand_dims increases ndim by introducing a new axis, while numpy.squeeze reduces ndim by removing axes of length 1. reshape can also alter ndim, provided the total number of elements remains consistent.

Notes and usage. numpy.ndim is convenient for generic code that accepts array-like inputs, while arr.ndim is specific to ndarray objects. Understanding ndim helps with indexing, broadcasting, and validating input shapes before performing operations.

Other libraries use similar concepts with different names. For instance, PyTorch tensors have a ndim attribute (or dim() in older code), while TensorFlow uses tf.rank to query dimensionality. In MATLAB and Octave, ndims serves a related purpose.