Support for OGB Hetero-Graphs

[Dsantra92]

Handling Hetero Graphs in OGBDatasets.

[codecov-commenter]

Codecov Report

Merging #140 (55291f0) into master (0fd7858) will increase coverage by 9.65%.
The diff coverage is 76.43%.

@@            Coverage Diff             @@
##           master     #140      +/-   ##
==========================================
+ Coverage   29.55%   39.21%   +9.65%     
==========================================
  Files          39       39              
  Lines        1543     1752     +209     
==========================================
+ Hits          456      687     +231     
+ Misses       1087     1065      -22     

Impacted Files	Coverage Δ
src/graph.jl	50.00% <55.88%> (+8.33%)	⬆️
src/datasets/graphs/ogbdataset.jl	57.57% <80.89%> (+56.36%)	⬆️
src/abstract_datasets.jl	22.58% <0.00%> (-0.76%)	⬇️
src/datasets/vision/cifar10.jl	2.40% <0.00%> (-0.03%)	⬇️
src/datasets/graphs/planetoid.jl	100.00% <0.00%> (ø)
...rc/datasets/vision/cifar10_reader/CIFAR10Reader.jl	0.00% <0.00%> (ø)
.../datasets/vision/cifar100_reader/CIFAR100Reader.jl	0.00% <0.00%> (ø)
src/datasets/text/udenglish.jl	56.86% <0.00%> (+0.86%)	⬆️
src/datasets/vision/mnist_reader/readimages.jl	56.66% <0.00%> (+1.49%)	⬆️
... and 4 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 0fd7858...55291f0. Read the comment docs.

[CarloLucibello]

Related to #139

[CarloLucibello]

what's the status of this? on which specific ogb dataset are you testing?

[Dsantra92]

Testing it on ogbn-mag. Edge data section and docs is left. Completing by EOD.

[Dsantra92]

Not adding training indexes in this pr. Training splits will be handled as per the torch issue. We also need more documentation for HeteroGraphs (and maybe a tutorial). Not sure why docs are failing :)

[CarloLucibello]

Is ogbn-mag big? can we add it to the test set? If not in the CI tests we should at least add it to the local tests

[Dsantra92]

Is ogbn-mag big? can we add it to the test set? If not in the CI tests we should at least add it to the local tests

Yes, it might just be big enough to fail the CI. Let's test if the CI breaks, and then fix it.

[CarloLucibello]

node_data should be such that node_data[node_type] is a dict or a named tuples. Similar applies to edge_data.