Scikit-network

-------------------

* Allow Python 3.7, by Thomas Bonald
* Fix input format for KMeans, issue 548 raised by sgerbe

-------------------

* Fix sampling for GraphSage, by Simon Delarue
* Fix leakage on GNNs, by Thomas Bonald and Simon Delarue
* Update tutorial on GNNs with node inference, by Thomas Bonald and Simon Delarue

-------------------

* Update Graph neural networks (e.g., add GraphSAGE), by Simon Delarue
* Clean data home folder (set one folder per dataset collection, NetSet, Konect, ...), by Thomas Bonald
* Improve classification by diffusion, setting -1 to unreached nodes, by Thomas Bonald
* Fix bug on modularity, raised by Alessandro (543)
* Clean up source distribution, by Nicholas Bollweg (544)
* Safe file extraction, by TrellixVulnTeam
* Fix bug on dendrogram cut, raised by Nina Sachdev (546)
* Add a function to aggregate a graph per label, by Thomas Bonald

-------------------

* Fix documentation

-------------------

* Drop Python 3.7
* Update NumPy and SciPy requirements
* Add graph neural networks, by Simon Delarue (533)
* Add fit_predict / fit_transform where appropriate, by Thomas Bonald
* Add Louvain hierarchical clustering (bottom-up), by Thomas Bonald
* Improve classification by diffusion (vectorial), by Thomas Bonald
* Add F1 scores for classification, by Thomas Bonald
* Add cosine similarity metric for embeddings, by Thomas Bonald
* Add acyclic test for undirected graphs, by Thomas Bonald
* Update algorithms to accept all sparse matrix formats of scipy, by Thomas Bonald

-------------------

* Add module on regression, by Thomas Bonald
* Add connected components for bipartite graphs, by Thomas Bonald
* Update functions for loading graphs, by Thomas Bonald
* Fix shuffling nodes in Louvain (issue 521), by Thomas Bonald
* Add radius and eccentricity metrics, by Henry Carscadden (522)
* Add new use case (recommendation), by Thomas Bonald