* added HGT DBLP example

* typo

* Merge PyG master (#52)

* Adding the Facebok Page-Page dataset

* type hints

* documentation CI

* py 3.8

* fix links

* fix links

* fail on warning

* fail on warning

* fix doc

Co-authored-by: benedekrozemberczki <benedek.rozemberczki@gmail.com>

* revert

* Fix Documentation Rendering (#51)

* fix doc rendering

* fix linting

* retrigger checks

* remove pytorch 1.7.0 legacy code (#50)

* Fix `copy.deepcopy` within lazy `nn.dense.Linear` (#44)

* fix deepcopy within lazy Linear

* fix merge

* assert exception

* example to doc

* resolve conflict

* resolve conflict

* Add Figure and Equation to `to_hetero` docstring (#60)

* add tex

* add svg + docstring

* typo

* added equation

* Message Passing Hooks (#53)

* add hooks

* docstring

* add docstring

* allow modification of inputs/output

* add test for modifying output

* add additional asserts for modifying output test

* Rename `HeteroData.get_edges` and `HeteroData.get_nodes` (#58)

* rename to_edges and to_nodes

* typo

* `HeteroConv` (#64)

* clean heteroconv

* init

* init

* clean up

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* fix documentation

* bipartite function

* fix test CI

* remove pillow version

* clean up for merge

* Merge PyG master (#69)

* renaming: PointConv to PointNetConv

* Fix a broken link in datasets/gdelt.py (#2800)

* fix test

* re-add batching of strings

* add quick start table

* gnn cheatsheet

* remove pillow version

Co-authored-by: Dongkwan Kim <todoaskit@gmail.com>

* re-merge

* add lazy column to GNN cheatsheet (#70)

* `to_hetero_with_bases(model)` (#63)

* update

* fix linting

* basisconv

* add ValueError

* to_hetero_with_bases impl done

* add test

* add comments

* add comments

* docstring

* typo

* update figure

* svg

* typo

* add test

* update

* add rgcn equality test

* typos

* update

* typos

* update figures

* generate new svgs

* fix assignment

* rename

* delete sorted edge types

* rename

* add legend

* fix typo

* Test: Check equal outputs of `to_hetero` and `RGCNConv` (#59)

* check equal output

* add sparsetensor test

* check equal output

* add sparsetensor test

* rename

* linting

* add missing import

* `HeteroData` support for `T.NormalizeFeatures` (#56)

* normalize features

* allow normalization of any feature

* in-place div

* normalize features

* allow normalization of any feature

* in-place div

* fix test

* no need to re-assign

* `HeteroData` support for `T.AddSelfLoops` (#54)

* hetero support for AddSelfLoops

* check for edge_index attribute

* f-string

* retrigger checks

* revert bipartite changes

* hetero support for AddSelfLoops

* check for edge_index attribute

* f-string

* retrigger checks

* revert bipartite changes

* merge master

* merge master

* `HeteroData` support for `T.ToSparseTensor` (#55)

* hetero support for ToSparseTensor

* add test

* customize the attribute of SparseTensor.value

* rework sort_edge_index

* hetero support for ToSparseTensor

* add test

* customize the attribute of SparseTensor.value

* rework sort_edge_index

* linting

* `HeteroData` support for `T.ToUndirected` (#57)

* to_undirected

* revert bipartite changes

* coalesce + undirected enhancement

* merge master

* revert bipartite changes

* coalesce + undirected enhancement

* merge master

* clean up

* new default relation type

* fix tests

* resolve merge conflicts

* resolve merge conflicts 2

* resolve merge conflicts 3

* Merge PyG master (#74)

* renaming: PointConv to PointNetConv

* Fix a broken link in datasets/gdelt.py (#2800)

* fix test

* re-add batching of strings

* add quick start table

* gnn cheatsheet

* remove pillow version

* clean up doc for to_dense_batch

* clean up

* add legend to cheatsheet

* Improve terminology (#2837)

I think the previous version of the document uses the term 'symmetric' incorrectly. A symmetric matrix is a square matrix that is is equal to its transpose (https://en.wikipedia.org/wiki/Symmetric_matrix). However, the text is only talking about the shape of the matrix, not its content. Hence, 'square (matrix)' would be the correct term to use.

* Add batch_size input to to_dense_batch (#2838)

* Add batch_size input to to_dense_batch

* to_dense_batch fix typo in batch_size param use

* add typehints

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* typo

* Added return_attention_weights to TransformerConv. (#2807)

* added return_weights functionality to tranformer

* added return attn weights tests

* flake8

* added typehints

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* MD17 (#2843)

* Added MD17 dataset

* Updated Documentation

* Added link to sGDML website in doc

* fixed typos in doc and made train variable description clearer

* clean up

* fix linting

* fix doc warning

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* update doc

* remove forward doc

* add static graph support info to cheatsheet

* fix num_nodes in case edge_index is empty

* fix math formula

* faster GDC import

* lazy import

* lazy import for datasets

* lazy import for nn

* Sequential jittable + traceable

* typo

* typo

* update doc

Co-authored-by: Dongkwan Kim <todoaskit@gmail.com>
Co-authored-by: Markus <markus.zopf@outlook.com>
Co-authored-by: Jimmie <jimmiebtlr@gmail.com>
Co-authored-by: Jinu Sunil <jinu.sunil@gmail.com>
Co-authored-by: Moritz R Schäfer <moritz.schaefer@protonmail.com>

* re-add

* GraphGym cleaned version (#82)

* GraphGym cleaned version

* remove deepsnap dependency

* fix lint errors, part 1

* fix all lint errors

* fix all lint errors

* fix all lint errors

* apply yapf

* Update .gitignore

* Integrate GraphGym into PyG (#85)

* GraphGym cleaned version

* remove deepsnap dependency

* fix lint errors, part 1

* fix all lint errors

* fix all lint errors

* fix all lint errors

* apply yapf

* Integrate graphgym into pyg, keep user API in project root

* fix merge conflict

* fix lint errors

* Make optional dependencies

* merge LICENSE from GraphGym

* add import

* clean up LICENSE

* fix import

* resolve merge conflicts

* resolve merge conflicts 2

* Merge PyG master (#87)

* renaming: PointConv to PointNetConv

* Fix a broken link in datasets/gdelt.py (#2800)

* fix test

* re-add batching of strings

* add quick start table

* gnn cheatsheet

* remove pillow version

* clean up doc for to_dense_batch

* clean up

* add legend to cheatsheet

* Improve terminology (#2837)

I think the previous version of the document uses the term 'symmetric' incorrectly. A symmetric matrix is a square matrix that is is equal to its transpose (https://en.wikipedia.org/wiki/Symmetric_matrix). However, the text is only talking about the shape of the matrix, not its content. Hence, 'square (matrix)' would be the correct term to use.

* Add batch_size input to to_dense_batch (#2838)

* Add batch_size input to to_dense_batch

* to_dense_batch fix typo in batch_size param use

* add typehints

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* typo

* Added return_attention_weights to TransformerConv. (#2807)

* added return_weights functionality to tranformer

* added return attn weights tests

* flake8

* added typehints

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* MD17 (#2843)

* Added MD17 dataset

* Updated Documentation

* Added link to sGDML website in doc

* fixed typos in doc and made train variable description clearer

* clean up

* fix linting

* fix doc warning

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* update doc

* remove forward doc

* add static graph support info to cheatsheet

* fix num_nodes in case edge_index is empty

* fix math formula

* faster GDC import

* lazy import

* lazy import for datasets

* lazy import for nn

* Sequential jittable + traceable

* typo

* typo

* update doc

* Simple models (#2869)

* Inclusion of new backbone models

* Eliminating head from asap.py

* small correction

* Create test_gcn.py

* Update __init__.py

* Update test_gcn.py

* Left only the convolutional simple models

* Tests included

* update

* clean up

* clean up v2

* fix activation

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* Example for MemPooling. (#2729)

* example for mem pooling

* backprop on kl loss is done at the end of an epoch. Keys in memory layers are trained only on kl loss.

* added learning rate decay. Using PROTIENS_full

* flake8

* reduced lr. increased weight decay

* changed download location

* added comments

* clean up

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* typos

* fix removeisolatednodes transform in case 'data.num_nodes' is present

* fix XConv with dilation > 1

* fix XConv with dilation > 1

* rgcn link prediction  (#2734)

* implemented LinkPrediction dataset for loading FB15k237

* implemented evaluation for relational link prediction

* implemented R-GCNConf link prediction example

* fixed bug: wrong initial objects in negative_sampling

* changed file downloader urllib.request.urlretrieve  to pytorch.data.download_url; renamed LinkPrediction class to RelationalLinkPredictionDataset

* update dataset

* update example script

* rename

Co-authored-by: Moritz <moritzblum>
Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* fix gnnexplainer draw kwargs

* remove python-louvain dependency

* allow customization of output in MP jit mode

* fix test for py3.6

* changed normalisation to same norm from instance norm to be robust to small var (#2917)

* add CITATION.cff

* format

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* [ci skip]

* add basetransform ABC (#2924)

* clean up BaseTransform

* clean up GATConv and add comments

* add max_num_neighbors as an additional argument

* fix jit GATConv on PyTorch 1.8.0

* fix doc

* fix gnn explainer with existing self-loops

* Rgcn link pred fix (#2946)

* added regularization, removed typo in test

* clean up

Co-authored-by: Moritz <moritzblum>
Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* typo

* Correct gini coefficient mathcal formula (#2932)

* typo

* typo

* Update from_networkx (#2923)

* Update from_networkx

* Update test

* Update convert.py

* Minor corrections

* Update test_convert.py

* Corrections

* Update test_convert.py

* Case where there are no edges

* Correcting how edge_attr are concatenated

* clean up + new test

* remove unused code

* add union type

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* fix deterministic ordering in from_networkx

* recursive-include *.jinja files

Co-authored-by: Dongkwan Kim <todoaskit@gmail.com>
Co-authored-by: Markus <markus.zopf@outlook.com>
Co-authored-by: Jimmie <jimmiebtlr@gmail.com>
Co-authored-by: Jinu Sunil <jinu.sunil@gmail.com>
Co-authored-by: Moritz R Schäfer <moritz.schaefer@protonmail.com>
Co-authored-by: PabloAMC <pmorenocf@alumnos.unex.es>
Co-authored-by: Moritz Blum <31183934+moritzblum@users.noreply.github.com>
Co-authored-by: fbragman <fbragman@users.noreply.github.com>
Co-authored-by: Christopher Lee <2824685+CCInc@users.noreply.github.com>
Co-authored-by: Tim Daubenschütz <tim@daubenschuetz.de>

* resolve merge conflicts 3

* resolve merge conflicts 4

* Implementation of the `HGTLoader` + `ogbn-mag` example (#73)

* first try

* update

* HGT Loader

* typo

* first try

* update

* HGT Loader

* typo

* bugfixes

* lazy GATConv

* bugfix

* bugfix

* full working pipeline

* update

* rename

* docstring

* typos

* update

* typo

* typo

* typo

* added comments

* add test

* add tests

* fix example

* rename

* linting

* Random split functionalities (#72)

* link split

* create split

* example tests

* link split tests

* fix linting

* update docstring

* undirected option, refactor and docs

* add num nodes as argument to neg sampling

* clean up + remove single object

* update example

* typo

* fix compose

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* add basetransform

* typo

* typo

* fix test

* Improve `torch_geometric.data` Documentation (#98)

* update data doc

* typo

* typo

* note

* typo

* add docstring

* only show inherited members for data and hetero_data

* documentation update for batch and dataset

* update doc

* update

* fix

* record_stream

* update

* typo

* add/fix data functionality

* linting

* typo

* `_parent` memory leak fix (#103)

* memory leak fix

* Clean up

* clean up

* bugfix tests

* typos

* fix test

* fix test

* rename reverse

* (Heterogeneous) `NeighborLoader` (#92)

* initial commit

* typo

* neighbor loader functionality + tests

* docstring

* fix docstring

* skip tests

* fix share_memory_

* typo

* typo

* update example

* typo

* share_strategy

* fix cuda calls

* better print

* fix size

* fix print

* final commit

* fix

* some todos

* preprocessed features

* fix to_undirected

* more documentation

* update doc

* fix doc

* fix doc

* Add benchmark code and the example with existing graph classification examples (#93)

* add benchmarking utilities

* update graph classification benchmark

* improve code style

* add pytorch-memlab for benchmark code

* skip some tests when cuda is not available

* add type hint when appropriate

* add seed_everything to improve code

* code refactoring

* code refactoring

* code refactoring

* code improvement

* remove unnecessary dataloader import

* change benchmark interface with decorator

* documentation improvement

* linting

* linting part 2

* linting part 3

* seed_everything

* create utils file

* update

* use utils functions

* fix test

* update the profiler to the latest torch (1.8.1+)

* refactor profiler and add more documentation

* refactor profiler and add more documentation

* resolve lint errors

* resolve lint errors

* update

* clean up test and profile

* fix linting

* add to doc

* fix doc

* typo

* update benchmark

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* Move `HGTLoader` to `torch_geometric.loader` + clean up (#104)

* move files

* use utils functions

* fix example

* update

* fix tests

* fix seed

* fix linear test

* rename

* Support GraphGym custom modules outside PyG package (#102)

* GraphGym cleaned version

* remove deepsnap dependency

* fix lint errors, part 1

* fix all lint errors

* fix all lint errors

* fix all lint errors

* apply yapf

* Integrate graphgym into pyg, keep user API in project root

* fix merge conflict

* fix lint errors

* Make optional dependencies

* merge LICENSE from GraphGym

* Enable adding GraphGym customized modules outside PyG package

* lint

* Rename `AddTrainValTestMask` to `RandomNodeSplit` (#108)

* initial commit

* rename example

* remove AddTrainValTestMask

* fix linting

* create optimizer config and scheduler config separately (#113)

* create optimizer config and scheduler config separately

* fix format

* import explicitly

Co-authored-by: Dong Wang <dongwang@yannis-air.lan>

* Heterogeneous Graph Tutorial (#83)

* add HG tutorial roadmap

* started working on hg tutorial

* hg_tutorial, some text and .tex figure

* added svg

* hg tutorial content

* fix CI

* text and structure

* finished first draft

* fixed one code example

* fixing conventions

* fixing links

* update svg

* some smaller improvements of tutorial

* improvements on tutorial

* hg-tutorial: fixed compiling issue, added detailed content

* added absolute links

* fixed warnings

* streamlined dataset section

* update svg

* update tutorial

* update 2

Co-authored-by: Jan Eric Lenssen <janeric.lenssen@tu-dortmund.de>

* typo

* Move data loaders to `torch_geometric.loader` (#110)

* move graphsaint

* deprecations

* move clusterloader

* deprecations

* type hints

* move shadow

* typo

* typo

* move datalistloader

* dense data loader

* random node sampler

* fix doc

* Lazy GNN operators (#89)

* lazy cheb conv

* lazy GraphConv

* lazy GATv2Conv

* lazy TAGConv

* lazy FAConv

* lazy FeaStConv

* lazy NNConv

* typo

* fix tests

* lazy SuperGATConv

* lazy SuperGATConv fix

* lazy SplineConv

* fix lazy check

* lazy GravNetConv

* arma conv lazy

* dense linear in gmmconv

* typo

* add test

* lazy GMMConv

* doc

* rename (#116)

* Revisit `MetaPath2Vec` (#114)

* revisit metapath2vec

* update

* typo

* update

* fix doc

* update

* check for attributes rather than key

* Clean up `torch_geometric.profile` further (#111)

* remove print_layer_stats

* typos

* update

* readme highlights and quick tour (#99)

* readme highlights and quick tour

* arch

* arch image

* arch overview

* list categories

* categorization

* category description

* Update README.md

from Matthias

Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* improved highlights

* Update README.md

Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* Update README.md

Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* Update README.md

Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* Update README.md

Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* minor

* update readme

* update

* update

* update

* update

* fix url

* update

* update

* update

* update

* update

* update

* move ops

* toc

* typo

* typo

* add svgs

* update figure

* fix links

* fix size

* fix size

* typo

Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* fix broken links

* fix links

* Heterogeneous Graph Sampler Tutorial (#117)

* initial commit

* address comments

* remove todo

* typo

* Conversion between heterogenous and homogeneous graph objects (#115)

* temp checkpoint (wip, will remove)

* (wip) typed graph conversion

* (wip) typed graph conversion

* (wip) typed graph conversion

* update

* typo

* delete examples

Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* fix test

* update doc

* deprecate NeighborSampler (#119)

* Move `torch_geometric.data.DataLoader` to `torch_geometric.loader.DataLoader` (#120)

* move dataloader

* rename

* typos

* typos

* fix __cat_dim__

* updategp

* Deprecate `train_test_split_edges` + Modifications to `RandomLinkSplit` (#121)

* deprecate train_test_split_edges

* to device transform

* fix example

* add split_labels argument

* fix autoencoder example

* typos

* add docstring

* ARGVA

* seal

* adress comments

* Create example to load `*.csv` and transfer to `HeteroData` (#76)

* create example to load csv file and transfer to heter-data

* add ipython notebook version load csv with documentation

* address comment

* first version of csv loading doc

* first version of csv loading doc

* suggestion docs/source/notes/loading_csv.rst

Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* suggestion docs/source/notes/loading_csv.rst

Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* suggestion docs/source/notes/loading_csv.rst

Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* suggestion docs/source/notes/loading_csv.rst

Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* suggestions csv tutorial

* example script load csv + extract fix

* fixed edge index stacking dimension in example and jupyter nb

* linting

* linting2

* rename

* update

* update

* update

* typo

* typo

* update

* rename

* update tutorial

* typo

* address comments

Co-authored-by: Dong Wang <dongwang@yannis-air.lan>
Co-authored-by: Jan Eric Lenssen <janeric.lenssen@tu-dortmund.de>
Co-authored-by: Matthias Fey <matthias.fey@tu-dortmund.de>

* typo

* fix

* typo

* update

* fix

* fix

Co-authored-by: benedekrozemberczki <benedek.rozemberczki@gmail.com>
Co-authored-by: Rex Ying <rexying@stanford.edu>
Co-authored-by: Dongkwan Kim <todoaskit@gmail.com>
Co-authored-by: Markus <markus.zopf@outlook.com>
Co-authored-by: Jimmie <jimmiebtlr@gmail.com>
Co-authored-by: Jinu Sunil <jinu.sunil@gmail.com>
Co-authored-by: Moritz R Schäfer <moritz.schaefer@protonmail.com>
Co-authored-by: Jiaxuan <youjiaxuan@gmail.com>
Co-authored-by: PabloAMC <pmorenocf@alumnos.unex.es>
Co-authored-by: Moritz Blum <31183934+moritzblum@users.noreply.github.com>
Co-authored-by: fbragman <fbragman@users.noreply.github.com>
Co-authored-by: Christopher Lee <2824685+CCInc@users.noreply.github.com>
Co-authored-by: Tim Daubenschütz <tim@daubenschuetz.de>
Co-authored-by: Yue Zhao <yzhao062@gmail.com>
Co-authored-by: Dong Wang <dongw89@gmail.com>
Co-authored-by: Dong Wang <dongwang@yannis-air.lan>
Co-authored-by: Jan Eric Lenssen <janeric.lenssen@tu-dortmund.de>