rustpyppr 0.1.2

No project description provided

Personalized PageRank in Rust for Python

This small library provides efficient functions for Python to compute Personalized PageRank scores in a graph. Personalized PageRank is similar to ordinary PageRank but the node where the random surfer will start from (after a teleportation or in the beginning) is not sampled uniformly from all the nodes, but rather from a smaller subset of them. This subset, which is called Personalized Set, must contain at least 1 node.

The implemented algorithms so far are various versions of Forward Push for single source. This means that given a source node, it is possible to compute the PPR scores of other ones w.r.t. the source node.

Getting Started

At the moment, we did not publish the wheel on pypi yet. After cloning this repository, you can manually build the wheel and install it.

pip install maturin
maturin build --release
pip install target/wheels/rustpyppr-[ wheel version and parameters here ]

After the installation, you can simply import it from python and use it.

import rustpyppr

graph = {3:[5, 1], 1:[3], 5:[3]}
source = 3
ppr = rustpyppr.forward_push(graph, source)
print(f'{ppr = }')
# ppr = {1: 0.22936248564736167, 3: 0.5412750287052767, 5: 0.22936248564736167}

# you can also run PPR on multiple sources in the same graph in parallel with native multithreading
sources = [3, 1]
ppr = rustpyppr.multiple_forward_push(graph, sources)
print(f'{ppr = }')
# ppr = {1: {1: 0.34579401171205487, 5: 0.19511406668236952, 3: 0.45909192160557544}, 3: {3: 0.5412750287052767, 5: 0.22936248564736167, 1: 0.22936248564736167}}

# and finally, you can run PPR on different graphs with different sources (also using multithreading)
graphs = [{2:[0], 0:[2]}, {3:[5, 1], 1:[3], 5:[3]}]
sources = [2, 5]
ppr = rustpyppr.parallel_forward_push(graphs, sources)
print(f'{ppr = }')
# ppr = {5: {3: 0.45909192160557544, 1: 0.19511406668236952, 5: 0.34579401171205487}, 2: {0: 0.4587249712947233, 2: 0.5412750287052767}}