image image image image image image image image image image image

Sparse vectors optimized for memory and NumPy integrations.

numpy handles densely populated n-dimemsional arrays. scipy.sparse handles sparsely populated 2-dimensional arrays, i.e., matrices. What’s missing from the ecosystem is sparsely populated 1-dimensional arrays, i.e., vectors.

NumPy | Python | Spector —– | —— | ——- 1-dim bool numpy.array | Set[int] | spector.indices 1-dim float numpy.array | Dict[int, float] | spector.vector scipy.sparse.dok_matrix | Dict[int, Dict[int, float]] | spector.matrix

Indices and vectors are implemented in Cython as hash sets and maps. All native operations are optimized and release the GIL.

• conversion between sparse numpy arrays
• conversion between dense numpy arrays
• binary set operations
• binary math operations
• map, filter, and reduce operations with numpy universal functions

Usage

indices

A sparse boolean array with a set interface.

>>> from spector import indices
>>> ind = indices([0, 2])
>>> ind
indices([2 0])
>>> 1 in ind
False
>>> ind.add(1)
True
>>> ind.todense()
array([ True,  True,  True])
>>> ind.fromdense(_)
indices([2 1 0])

vector

A sparse float array with a mapping interface.

>>> from spector import vector
>>> vec = vector({0: 1.0, 2: 2.0, 4: 1.0})
>>> vec
vector([4 2 0], [1. 2. 1.])
>>> vec[2] += 1.0
>>> vec[2]
3.0
>>> vec.todense()
array([1., 0., 3., 0., 1.])
>>> vector.fromdense(_)
vector([4 2 0], [1. 3. 1.])
>>> vec.sum()
5.0
>>> vec + vec
vector([0 2 4], [2. 6. 2.])

Vectors support math operations with scalars, and with vectors if the set method is unambiguous.

vector operation | set method | ufunc —————- | ———- | —– + | union | add * | intersection | multiply - | | subtract / | | true_divide ** | | power \| | union | max & | intersection | min ^ | symmetric_difference | difference | difference |

matrix

A mapping of keys to vectors.

>>> from spector import matrix
>>> mat = matrix({0: {1: 2.0}})
>>> mat
matrix(<class 'spector.vector.vector'>, {0: vector([1], [2.])})
>>> mat.row, mat.col, mat.data
(array([0]), array([1]), array([2.]))

Installation

% pip install spector

Tests

100% branch coverage.

% pytest [--cov]

Changes

dev

• Python >=3.6 required

1.0

• Optimized set intersection

0.2

• Numerous optimizations
• Intersection count
• Increased parallelism