This package contains algorithms for graph embedding for networks. All algorithms are implemented in Python, with speed accelerations by numba.

• Installation
• Examples
• Available Algorithms

pip install git+https://github.com/skojaku/graphvec.git

For conda users,

git clone https://github.com/skojaku/graphvec
cd graphvec
conda develop .

See setup.py

Following the scikit-learn's API design, all algorithms have two methods.fit and .transform. The .fit method takes a network to learn the network structure, and the .transform produces the emebdding of nodes. More specifically, follow the steps below.

First, load a graph embedding method. For demonstration, we use Node2Vec. See the "Available algorithms" Section for other graph embedding methods.

import graphvec
model = graphvec.Node2Vec()

Second, call .fit with the network A:

model.fit(A)

where A is the adjacency matrix in scipy.sparse format. For networkx user, networkx.Graph can be converted to the scipy.sparse format by A = nx.adjacency_matrix(G).

Lastly, call .transform to generate an embedding:

emb = model.transform(dim=64)

where dim is the number of dimensions. emb is a numpy.ndarray with size (number of nodes, dim). The ith row in emb (emb[i, :]) is the embedding vector of the ith node in the given adjacency matrix.

conda create -n graphvec python=3.9
conda activate graphvec
conda install -c conda-forge mamba -y
mamba install -c conda-forge graph-tool scikit-learn numpy numba scipy pandas networkx seaborn matplotlib gensim ipykernel tqdm black -y
mamba install -c conda-forge numpy==1.23.5