Feature/attri2vec

README.md

@@ -73,7 +73,7 @@ The StellarGraph library can be used to solve tasks using graph-structured data,
 - Representation learning for nodes and edges, to be used for visualisation and various downstream machine learning tasks;
 - Classification and attribute inference of nodes or edges;
 - Link prediction;
-- Interpretation of node classification through calculated importances of edges and neighbours for selected nodes [7].
+- Interpretation of node classification through calculated importances of edges and neighbours for selected nodes [8].
 
 We provide [examples](https://github.com/stellargraph/stellargraph/tree/master/demos/) of using `StellarGraph` to solve such tasks using several real-world datasets.
 
@@ -130,7 +130,7 @@ can be downloaded and installed from [python.org](https://python.org/). Alternat
 environment, available from [anaconda.com](https://www.anaconda.com/download/).
 
 *Note*: while the library works on Python 3.7 it is based on Keras which does not officially support Python 3.7.
-Therefore, there may be unforseen bugs and you there are many warnings from the Python libraries that 
+Therefore, there may be unforseen bugs and you there are many warnings from the Python libraries that
 StellarGraph depends upon.
 
 <!--
@@ -172,7 +172,7 @@ pip install .[demos]
 
 * [stellargraph/stellargraph](https://hub.docker.com/r/stellargraph/stellargraph): Docker image with `stellargraph` installed.
 
-Images can be pulled via `docker pull stellargraph/stellargraph` 
+Images can be pulled via `docker pull stellargraph/stellargraph`
 
 
 ## Running the examples
@@ -193,13 +193,16 @@ The StellarGraph library currently includes the following algorithms for graph m
   The current implementation supports mean aggregation of neighbour nodes,
   taking into account their types and the types of links between them.
 
-* The Graph ATtention Network (GAT) [4]
+* Attri2vec [4]
+  - Supports node representation learning, node classification, and out-of-sample node link prediction for homogeneous graphs with node attributes.
+
+* The Graph ATtention Network (GAT) [5]
   - The GAT algorithm supports representation learning and node classification for homogeneous graphs. There are versions of the graph attention layer that support both sparse and dense adjacency matrices.
 
-* Graph Convolutional Network (GCN) [5]
+* Graph Convolutional Network (GCN) [6]
   - The GCN algorithm supports representation learning and node classification for homogeneous graphs. There are versions of the graph convolutional layer that support both sparse and dense adjacency matrices.
 
-* Simplified Graph Convolutional network (SGC) [6]
+* Simplified Graph Convolutional network (SGC) [7]
   - The SGC network algorithm supports representation learning and node classification for homogeneous graphs. It is an extension of the GCN algorithm that smooths the graph to bring in more distant neighbours of nodes without using multiple layers.
 
 * Node2Vec [2]
@@ -262,14 +265,17 @@ Neural Information Processing Systems (NIPS), 2017. ([link](https://arxiv.org/ab
 ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 135–144, 2017
 ([link](https://ericdongyx.github.io/metapath2vec/m2v.html))
 
-4. Graph Attention Networks. P. Velickovic et al.
+4. Attributed Network Embedding via Subspace Discovery. D. Zhang, Y. Jie, X. Zhu and C. Zhang, arXiv:1901.04095,
+[cs.SI], 2019. ([link](https://arxiv.org/abs/1901.04095))
+
+5. Graph Attention Networks. P. Velickovic et al.
 International Conference on Learning Representations (ICLR) 2018 ([link](https://arxiv.org/abs/1710.10903))
 
-5. Graph Convolutional Networks (GCN): Semi-Supervised Classification with Graph Convolutional Networks. Thomas N. Kipf, Max Welling.
+6. Graph Convolutional Networks (GCN): Semi-Supervised Classification with Graph Convolutional Networks. Thomas N. Kipf, Max Welling.
 International Conference on Learning Representations (ICLR), 2017
 ([link](https://github.com/tkipf/gcn))
 
-6. Simplifying Graph Convolutional Networks. F. Wu, T. Zhang, A. H. de Souza, C. Fifty, T. Yu, and K. Q. Weinberger. 
+7. Simplifying Graph Convolutional Networks. F. Wu, T. Zhang, A. H. de Souza, C. Fifty, T. Yu, and K. Q. Weinberger.
 International Conference on Machine Learning (ICML), 2019. ([link](https://arxiv.org/abs/1902.07153))
 
-7. Adversarial Examples on Graph Data: Deep Insights into Attack and Defense. H. Wu, C. Wang, Y. Tyshetskiy, A. Docherty, K. Lu, and L. Zhu. IJCAI 2019. ([link](https://arxiv.org/abs/1903.01610))
+8. Adversarial Examples on Graph Data: Deep Insights into Attack and Defense. H. Wu, C. Wang, Y. Tyshetskiy, A. Docherty, K. Lu, and L. Zhu. IJCAI 2019. ([link](https://arxiv.org/abs/1903.01610))

demos/README.md

@@ -7,10 +7,10 @@ examples demonstrate using the `StellarGraph` library to build machine learning
 workflows on both homogeneous and heterogeneous networks.
 
 Each folder contains one or more examples of using the StellarGraph implementations of the
-state-of-the-art algorithms, GraphSAGE [3], HinSAGE, GCN [5], GAT [6], Node2Vec [1], and Metapath2Vec [2]. 
-GraphSAGE, HinSAGE, and GAT are variants of Graph Convolutional Neural networks [5]. Node2Vec and
+state-of-the-art algorithms, attri2vec [4], GraphSAGE [3], HinSAGE, GCN [6], GAT [7], Node2Vec [1], and Metapath2Vec [2].
+GraphSAGE, HinSAGE, and GAT are variants of Graph Convolutional Neural networks [6]. Node2Vec and
 Metapath2Vec are methods based on graph random walks and representation learning using the
-Word2Vec [4] algorithm.
+Word2Vec [5] algorithm.
 
 The examples folder structure is shown below.
 
@@ -30,8 +30,8 @@ The examples folder structure is shown below.
 * [`/ensembles`](https://github.com/stellargraph/stellargraph/tree/master/demos/ensembles)
 
     Examples of using ensembles of graph convolutional neural networks, e.g., GraphSAGE, GCN, HinSAGE, etc., for
-    node classification and link prediction. Model ensembles usually yield better predictions than single models, 
-    while also providing estimates of prediction uncertainty as a bonus. 
+    node classification and link prediction. Model ensembles usually yield better predictions than single models,
+    while also providing estimates of prediction uncertainty as a bonus.
 
 * [`/calibration`](https://github.com/stellargraph/stellargraph/tree/master/demos/calibration)
 
@@ -49,24 +49,27 @@ The examples folder structure is shown below.
 
 ## References
 
-1. Node2Vec: Scalable Feature Learning for Networks. A. Grover, J. Leskovec. ACM SIGKDD International Conference on 
+1. Node2Vec: Scalable Feature Learning for Networks. A. Grover, J. Leskovec. ACM SIGKDD International Conference on
 Knowledge Discovery and Data Mining (KDD), 2016. ([link](https://snap.stanford.edu/node2vec/))
 
-2. Metapath2Vec: Scalable Representation Learning for Heterogeneous Networks. Yuxiao Dong, Nitesh V. Chawla, and 
+2. Metapath2Vec: Scalable Representation Learning for Heterogeneous Networks. Yuxiao Dong, Nitesh V. Chawla, and
 Ananthram Swami. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 135–144, 2017
 ([link](https://ericdongyx.github.io/metapath2vec/m2v.html))
 
-3. Inductive Representation Learning on Large Graphs. W.L. Hamilton, R. Ying, and J. Leskovec arXiv:1706.02216 
+3. Inductive Representation Learning on Large Graphs. W.L. Hamilton, R. Ying, and J. Leskovec arXiv:1706.02216
 [cs.SI], 2017. ([link](http://snap.stanford.edu/graphsage/))
 
-4. Distributed representations of words and phrases and their compositionality. T. Mikolov, 
+4. Attributed Network Embedding via Subspace Discovery. D. Zhang, Y. Jie, X. Zhu and C. Zhang, arXiv:1901.04095,
+[cs.SI], 2019. ([link](https://arxiv.org/abs/1901.04095))
+
+5. Distributed representations of words and phrases and their compositionality. T. Mikolov,
 I. Sutskever, K. Chen, G. S. Corrado, and J. Dean. In Advances in Neural Information Processing
  Systems (NIPS), pp. 3111-3119, 2013. ([link](https://papers.nips.cc/paper/5021-distributed-representations-of-words-and-phrases-and-their-compositionality.pdf))
 
-5. Semi-Supervised Classification with Graph Convolutional Networks. T. Kipf, M. Welling. 
+6. Semi-Supervised Classification with Graph Convolutional Networks. T. Kipf, M. Welling.
 ICLR 2017. arXiv:1609.02907 ([link](https://arxiv.org/abs/1609.02907))
 
-6. Graph Attention Networks. P. Velickovic et al. ICLR 2018 ([link](https://arxiv.org/abs/1710.10903))
+7. Graph Attention Networks. P. Velickovic et al. ICLR 2018 ([link](https://arxiv.org/abs/1710.10903))
 
-7. On Calibration of Modern Neural Networks. C. Guo, G. Pleiss, Y. Sun, and K. Q. Weinberger. 
+8. On Calibration of Modern Neural Networks. C. Guo, G. Pleiss, Y. Sun, and K. Q. Weinberger.
 ICML 2017. ([link](https://geoffpleiss.com/nn_calibration))

demos/embeddings/README.md

@@ -1,34 +1,40 @@
 ## Representation Learning Examples
 
-This folder contains three [Jupyter](http://jupyter.org/) python notebooks demonstrating the use of unsupervised graph representation learning methods implemented in the `stellargraph` library for homogeneous and hetrogenous graphs with or without node features. The original works are referenced below. 
+This folder contains three [Jupyter](http://jupyter.org/) python notebooks demonstrating the use of unsupervised graph representation learning methods implemented in the `stellargraph` library for homogeneous and hetrogenous graphs with or without node features. The original works are referenced below.
 
-**Node2Vec** and **Metapath2Vec** notebooks demonstrate the combined use of `stellargraph` and `Gensim` [4] libraries for representation learning on homogeneous and heterogeneous graphs. 
+**Node2Vec** and **Metapath2Vec** notebooks demonstrate the combined use of `stellargraph` and `Gensim` [4] libraries for representation learning on homogeneous and heterogeneous graphs.
 **Unsupervised GraphSAGE** notebook demonstrates the use of `Stellargraph` library's GraphSAGE implementation for unsupervised learning of node embeddings for homogeneous graphs with node features.
+**attri2vec** notebook demonstrates the implementation of attri2vec with the `Stellargraph` library for unsupervised inductive learning of node embeddings for homogeneous graphs with node features, and the evaluation for its ability to infer the representations of out-of-sample nodes with the out-of-sample node link prediction task.   
 
 The notebooks demonstrate the following algorithms.
 - `stellargraph-node2vec.ipynb` The **Node2Vec** algorithm [1] for representation learning on homogeneous graphs
 - `stellargraph-metapath2vec.ipynb` The **Metapath2Vec** algorithm [2] for representation learning on heterogeneous graphs.
 - `embeddings-unsupervised-graphsage-cora.ipynb` The **Unsupervised GraphSAGE** algorithm [5] for representation learning on homogeneous graphs with node features.
+- `stellargraph-attri2vec-citeseer.ipynb` The **attri2vec** algorithm [6] for representation learning on the homogeneous graph citeseer with node features.
+- `stellargraph-attri2vec-DBLP.ipynb` The **attri2vec** algorithm [6] for representation learning on the homogeneous graph DBLP with node features.
 
-All examples demonstrate how to calculate embedding vectors for a graph's nodes in just a few lines of Python code. 
+All examples demonstrate how to calculate embedding vectors for a graph's nodes in just a few lines of Python code.
 The learned node representations can be used in numerous downstream tasks such as node attribute inference, link
 prediction, and community detection.
 
 
 ## References
 
-**1.** Node2Vec: Scalable Feature Learning for Networks. A. Grover, J. Leskovec. ACM SIGKDD International Conference 
+**1.** Node2Vec: Scalable Feature Learning for Networks. A. Grover, J. Leskovec. ACM SIGKDD International Conference
 on Knowledge Discovery and Data Mining (KDD), 2016. ([link](https://snap.stanford.edu/node2vec/))
 
-**2.**  Metapath2Vec: Scalable Representation Learning for Heterogeneous Networks. Yuxiao Dong, Nitesh V. Chawla, and 
-Ananthram Swami. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 135–144, 2017. 
+**2.**  Metapath2Vec: Scalable Representation Learning for Heterogeneous Networks. Yuxiao Dong, Nitesh V. Chawla, and
+Ananthram Swami. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 135–144, 2017.
 ([link](https://ericdongyx.github.io/papers/KDD17-dong-chawla-swami-metapath2vec.pdf))
 
-**3.** Distributed representations of words and phrases and their compositionality. T. Mikolov, I. Sutskever, K. Chen, 
-G. S. Corrado, and J. Dean.  In Advances in Neural Information Processing Systems (NIPS), pp. 3111-3119, 2013. 
+**3.** Distributed representations of words and phrases and their compositionality. T. Mikolov, I. Sutskever, K. Chen,
+G. S. Corrado, and J. Dean.  In Advances in Neural Information Processing Systems (NIPS), pp. 3111-3119, 2013.
 ([link](https://papers.nips.cc/paper/5021-distributed-representations-of-words-and-phrases-and-their-compositionality.pdf))
 
 **4.** Gensim: Topic modelling for humans. ([link](https://radimrehurek.com/gensim/))
 
 **5.** Inductive Representation Learning on Large Graphs. W.L. Hamilton, R. Ying, and J. Leskovec arXiv:1706.02216
 [cs.SI], 2017. ([link](http://snap.stanford.edu/graphsage/))
+
+**6.** Attributed Network Embedding via Subspace Discovery. D. Zhang, Y. Jie, X. Zhu and C. Zhang, arXiv:1901.04095,
+[cs.SI], 2019. ([link](https://arxiv.org/abs/1901.04095))