Merge branch 'develop' of https://github.com/stellargraph/stellargraph …

…into feature/gat

.buildkite/pipeline.yml

@@ -1,25 +1,22 @@
 steps:
   - label: ":python:"
-    command:
-      - pip install -q --no-cache-dir -r requirements.txt -e .
-        && py.test -ra --cov=stellargraph tests/ --doctest-modules --doctest-modules --cov-report=term-missing -p no:cacheprovider
-        && coveralls
+    command: ".buildkite/script.sh"
     plugins:
-      docker#v1.3.0:
+      docker#v3.0.1:
         image: "python:3.6"
-        workdir: /app
         environment:
           - PYTHONDONTWRITEBYTECODE=1
-          - COVERALLS_REPO_TOKEN=$COVERALLS_REPO_TOKEN
+          - GIT_BRANCH=$BUILDKITE_BRANCH
+          - COVERALLS_REPO_TOKEN
+          - BUILDKITE_BUILD_NUMBER
+          - BUILDKITE_BRANCH
+          - AWS_LOGS_BUCKET
 
   - label: "style"
-    command:
-      - --check stellargraph tests
     plugins:
-      docker#v1.3.0:
+      docker#v3.0.1:
         image: "stellargraph/black"
-        workdir: /app
-        entrypoint: /usr/bin/black
+        command: ["--check", "stellargraph", "tests"]
         environment:
           - PYTHONDONTWRITEBYTECODE=1
 

.buildkite/script.sh

@@ -0,0 +1,15 @@
+#!/bin/bash
+
+set -xeo pipefail
+
+upload_tests() {
+  buildkite-agent artifact upload "${BUILDKITE_BUILD_NUMBER}.xml" s3://${AWS_LOGS_BUCKET}/pytest/${BUILDKITE_BRANCH}/${BUILDKITE_BUILD_NUMBER}
+}
+
+pip install -q --no-cache-dir -r requirements.txt -e .
+py.test -ra --cov=stellargraph tests/ --doctest-modules --doctest-modules --cov-report=term-missing -p no:cacheprovider --junitxml=./${BUILDKITE_BUILD_NUMBER}.xml
+coveralls
+
+if [ "${BUILDKITE_BRANCH}" = "develop"  ] || [ "${BUILDKITE_BRANCH}" = "master" ] ; then
+	upload_tests
+fi

.github/issue_template.md

@@ -16,6 +16,7 @@ so that:
 - [ ] Code well commented
 - [ ] Documentation on Google Docs
 - [ ] Documentation in repo
+- [ ] CHANGELOG.md updated
 - [ ] Team demo
 - [ ] Mini-meetup talk
 - [ ] Stakeholder sign-off
@@ -34,4 +35,6 @@ so that:
 - [ ] Branch and Pull Request pass integration tests on CI
 - [ ] Version number reflects new status
 - [ ] Peer Code Review Performed
-- [ ] Code well commented
+- [ ] Code well commented
+- [ ] CHANGELOG.md updated
+

.readthedocs.yaml

@@ -0,0 +1,10 @@
+# .readthedocs.yml
+
+version: 2
+
+python:
+  version: 3.6
+  install:
+     - requirements: requirements.txt
+     - method: pip
+       path: .

AUTHORS

@@ -14,6 +14,9 @@ GraphSAGE algorithm:
 Copyright (c) 2017 William L. Hamilton, Rex Ying | MIT License
 Copyright (c) 2016 Thomas Kipf | MIT License
 
+Graph Convolutional Networks (GCN) algorithm:
+Copyright (c) 2016 Thomas Kipf | MIT License
+
 GAT algorithm:
 Copyright (c) 2018 Daniele Grattarola | MIT License
 Copyright (c) 2018 Petar Veličković | MIT License

CHANGELOG.md

@@ -0,0 +1,31 @@
+# Change Log
+
+## [0.5.0](https://github.com/stellargraph/stellargraph/tree/v0.5.0) (11 Feb 2019)
+
+**Implemented new features and enhancements:**
+
+- Added model calibration [\#326](https://github.com/stellargraph/stellargraph/pull/326)
+- Added `GraphConvolution` layer, `GCN` class for a stack of `GraphConvolution` layers, 
+and `FullBatchNodeGenerator` class for feeding data into `GCN` models [\#318](https://github.com/stellargraph/stellargraph/pull/318)
+- Added GraphSAGE attention aggregator [\#317](https://github.com/stellargraph/stellargraph/pull/317)
+- Added GraphSAGE MaxPoolAggregator and MeanPoolAggregator [\#278](https://github.com/stellargraph/stellargraph/pull/278)
+- Added shuffle option to all `flow` methods for GraphSAGE and HinSAGE generators [\#328](https://github.com/stellargraph/stellargraph/pull/328)
+- GraphSAGE and HinSAGE: ensure that a MLP can be created by using zero samples [\#301](https://github.com/stellargraph/stellargraph/issues/301)
+- Handle isolated nodes in GraphSAGE [\#294](https://github.com/stellargraph/stellargraph/issues/294)
+- Ensure isolated nodes are handled correctly by GraphSAGENodeMapper and GraphSAGELinkMapper [\#182](https://github.com/stellargraph/stellargraph/issues/182)
+- EdgeSplitter: introduce a switch for keeping the reduced graph connected [\#285](https://github.com/stellargraph/stellargraph/issues/285)
+- Node2vec for weighted graphs [\#241](https://github.com/stellargraph/stellargraph/issues/241)
+- Fix edge types in demos [\#237](https://github.com/stellargraph/stellargraph/issues/237)
+- Add docstrings to StellarGraphBase class [\#175](https://github.com/stellargraph/stellargraph/issues/175)
+- Make L2-normalisation of the final embeddings in GraphSAGE and HinSAGE optional [\#115](https://github.com/stellargraph/stellargraph/issues/115)
+- Check/change the GraphSAGE mapper's behaviour for isolated nodes [\#100](https://github.com/stellargraph/stellargraph/issues/100)
+- Added GraphSAGE node embedding extraction and visualisation [\#290](https://github.com/stellargraph/stellargraph/pull/290)
+
+**Fixed bugs:**
+
+- Fixed the bug in running demos when no options given [\#271](https://github.com/stellargraph/stellargraph/issues/271)
+- Fixed the bug in LinkSequence that threw an error when no link targets were given [\#273](https://github.com/stellargraph/stellargraph/pull/273)
+
+**Refactoring:**
+- Refactored link inference classes to use `edge_embedding_method` instead of `edge_feature_method` [\#327](https://github.com/stellargraph/stellargraph/pull/327)
+

CONTRIBUTORS

@@ -1,5 +1,5 @@
 # The AUTHORS file lists the copyright holders; this file
-# lists people who contributed to StellarGraph codebase.  
+# lists people who contributed to StellarGraph codebase.
 # For example, CSIRO employees are listed here
 # but not in AUTHORS, because CSIRO holds the copyright.
 

README.md

@@ -136,7 +136,8 @@ The StellarGraph library currently includes the following algorithms for graph m
 
 * GraphSAGE [1]
   - Supports representation learning, node classification/regression, and link prediction for homogeneous networks.
-  The current implementation supports mean aggregation of neighbour nodes only.
+  The current implementation supports multiple aggregation methods, including mean, maxpool, meanpool, and
+  attentional aggregators.
 
 * HinSAGE
   - Extension of GraphSAGE algorithm to heterogeneous networks.
@@ -145,7 +146,12 @@ The StellarGraph library currently includes the following algorithms for graph m
   taking into account their types and the types of links between them.
 
 * GAT
-  - Graph ATtention Network algorithm [4] for homogeneous networks.
+  - Graph ATtention Network algorithm [4] for homogeneous graphs.
+
+* GCN
+  - Graph Convolutional Network algorithm [5] for homogeneous graphs. 
+  The implementation supports representation learning and node classification for homogeneous graphs.
+
 
 * Node2Vec [2]
   - Unsupervised representation learning for homogeneous networks, taking into account network structure while ignoring
@@ -191,4 +197,8 @@ Knowledge Discovery and Data Mining (KDD), 2016. ([link](https://snap.stanford.e
 Ananthram Swami. 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. ICLR 2018 ([link](https://arxiv.org/abs/1710.10903))
+4. Graph Attention Networks. P. Velickovic et al. 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.
+International Conference on Learning Representations (ICLR), 2017
+([link](https://github.com/tkipf/gcn))

demos/README.md

@@ -1,57 +1,66 @@
 ## Table of Contents
 
 This folder contains several examples of applying graph machine learning (ML) algorithms on network-structured
-data to solve several common problems including node attribute inference (inferring 
+data to solve several common problems including node attribute inference (inferring
 entity properties) and link prediction (predicting relations and/or their properties). The
-examples demonstrate using the `StellarGraph` library to build machine learning 
+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, Node2Vec [1], and Metapath2Vec [2]. 
-GraphSAGE and HinSAGE are variants of Graph Convolutional Neural networks [5]. Node2Vec and
+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
 Metapath2Vec are methods based on graph random walks and representation learning using the
 Word2Vec [4] algorithm.
 
-The examples folder structure is shown below. 
+The examples folder structure is shown below.
 
 * [`/embeddings`](https://github.com/stellargraph/stellargraph/tree/master/demos/embeddings)
 
-    Examples of unsupervised node representation learning for homogeneous and heterogeneous networks, 
+    Examples of unsupervised node representation learning for homogeneous and heterogeneous networks,
     using the Node2Vec and Metapath2Vec algorithms.
 
 * [`/link-prediction-random-walks`](https://github.com/stellargraph/stellargraph/tree/master/demos/link-prediction-random-walks)
 
-    Examples of semi-supervised link prediction for homogeneous and heterogeneous networks, 
+    Examples of semi-supervised link prediction for homogeneous and heterogeneous networks,
     using the Node2Vec and Metapath2vec algorithms.
-    
+
 * [`/link-prediction-graphsage`](https://github.com/stellargraph/stellargraph/tree/master/demos/link-prediction-graphsage)
 
-    Example of semi-supervised link prediction for a homogeneous network with attributed nodes, 
+    Example of semi-supervised link prediction for a homogeneous network with attributed nodes,
     using the GraphSAGE algorithm.
-    
+
 * [`/link-prediction-hinsage`](https://github.com/stellargraph/stellargraph/tree/master/demos/link-prediction-hinsage)
 
-    Example of supervised link attribute prediction for a heterogeneous network with attributed nodes of different types, 
+    Example of supervised link attribute prediction for a heterogeneous network with attributed nodes of different types,
     using the HinSAGE algorithm.
-    
+
 * [`/node-classification-graphsage`](https://github.com/stellargraph/stellargraph/tree/master/demos/node-classification-graphsage)
 
     Example of supervised node classification for a homogeneous network with attributed nodes, using the GraphSAGE algorithm.
 
+* [`/node-classification-gcn`](https://github.com/stellargraph/stellargraph/tree/master/demos/node-classification-gcn)
+
+    Example of semi-supervised node classification for a homogeneous network, using the GCN algorithm.
+
 * [`/node-classification-gat`](https://github.com/stellargraph/stellargraph/tree/master/demos/node-classification-gat)
 
     Example of supervised node classification for a homogeneous network with attributed nodes, using the GAT algorithm [6].
 
 * [`/node-classification-node2vec`](https://github.com/stellargraph/stellargraph/tree/master/demos/node-classification-node2vec)
 
-    Example of unsupervised node representation learning using Node2Vec and supervised classification using 
+    Example of unsupervised node representation learning using Node2Vec and supervised classification using
     the Scikit-learn library.
 
 * [`/node-classification-hinsage`](https://github.com/stellargraph/stellargraph/tree/master/demos/node-classification-hinsage)
 
-    Example of semi-supervised node classification for a heterogeneous network with multiple node and link types, 
+    Example of semi-supervised node classification for a heterogeneous network with multiple node and link types,
     using the HinSAGE algorithm.
 
+* [`/calibration`](https://github.com/stellargraph/stellargraph/tree/master/demos/calibration)
+
+    Examples of calibrating graph convolutional neural networks, e.g., GraphSAGE, for binary and
+    multi-class classification problems.
+
 
 ## References
 
@@ -73,3 +82,6 @@ I. Sutskever, K. Chen, G. S. Corrado, and J. Dean. In Advances in Neural Informa
 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. On Calibration of Modern Neural Networks. C. Guo, G. Pleiss, Y. Sun, and K. Q. Weinberger. 
+ICML 2017. ([link](https://geoffpleiss.com/nn_calibration))