Norm Conventions

DataVec.norm and DataMat.norm follow NumPy's conventions while preserving labels.

Vector Norms (DataVec.norm)

• ord=2 (default): Euclidean norm (|v|_2 = \sqrt{\sum_i v_i^2}).
• ord=1: Manhattan norm (\sum_i |v_i|).
• ord=\infty: maximum absolute entry.
• ord=0: count of non-zero elements.
• Negative orders and other values behave exactly like numpy.linalg.norm.

Matrix Norms (DataMat.norm)

• ord=None or 'fro': Frobenius norm (|A|F = \sqrt{\sum{i,j} a_{ij}^2}).
• ord=1: induced column-sum norm (\max_j \sum_i |a_{ij}|).
• ord=\infty: induced row-sum norm (\max_i \sum_j |a_{ij}|).
• ord=2: spectral norm (largest singular value).
• ord='nuc': nuclear norm (sum of singular values).
• Any other options are forwarded to numpy.linalg.norm.

Remember that calling these methods returns scalars; the underlying labels remain available on the original objects.