This is a patch release including the following bug fixes and new models.
Documentation fixes
- #2172 CoraFull dataset remove redundant reference
- #2167 #2177 Fix multiple docstring typos and inconsistencies in
dgl.dataloading
and minibatch training user guide. (Thanks @ustchhy for reviewing) - #2131 Update Doc for UDFs
Bug fixes
- #2128 cannot request out_edges() for empty node sets on cuda
- #2098 Context Issue for bfs_edges_generator on GPU Graphs
- #2135 Pickling a subgraph stores the feature of the original graph
- #2137 0.5.x taking too much shared memory during multiprocess training
- #2145 dgl.batch() in 0.5.x is slower than 0.4.x
- #2157 edge_softmax function not working on subgraphs
- #2161 #2165 #2173 TUDataset (Thanks @henrykenlay)
- #2175 Messages not ordered by edge IDs in degree bucketing
- #2166 Error when call apply_edges for dec_graph
- #2169 Multiprocessing neighbor sampling sometimes have the memory corrupted
- #2106 Bad file descriptor error when saving dgl graph to HDFS
- #2188 Fix dtype mismatch in EdgeDataLoader on Windows
Bug fixes in examples
- #2143 Fix unsupervised graphsage
- #2182 Use DistDataLoader instead of Pytorch’s DataLoader in Distributed GraphSAGE (Thanks @liucw2012 )
- #2187 Fix partition for 0.5.1
New examples
- #2153 GCN on OGB-Arxiv (Thanks @Espylapiza )