tgn
tgn copied to clipboard
twitter-research
TGN: Temporal Graph Networks
Hello, I have a couple of questions about the [plot](https://miro.medium.com/max/504/0*gmXGRalaeLMpQKNm) from Bronsteins et al.'s "Accelerating and scaling Temporal Graph Networks on the Graphcore IPU" article: 1- How did you separate...
Hi, let's say we are using GraphSumEmbedding as a layer in TGN. As I understood from [Your paper](https://arxiv.org/pdf/2006.10637.pdf) in the propagation part, we are supposed to concatenate embeddings from the...
In the preprocess code you currently only use a zero matrix, so it does not matter in your experiments. https://github.com/twitter-research/tgn/blob/2aa295a1f182137a6ad56328b43cb3d8223cae77/utils/preprocess_data.py#L74 But why is the feature matrix one row larger than...
return NotImplemented for some not implememented functions
Hi @emalgorithm, I got some problems when reading your codes. When `memory_update_at_start=True`, the msg_agg and msg_func will calc twice, before the `compute_embedding` and after `compute_embedding`. Before the `compute_embedding`, the `get_updated_memory`...
Hi team, when I tried to run script using Reddit dataset, it failed after 6 training epoch.  I run the script on NVIDIA RTX 3080ti, which does not support...
Hi, If I am not mistaken, there seems to be a bug when using the model on a Unipartite dataset when updating the memory at the end of each batch...
This pull request fixes issue #29. Instead of updating the memory separately for sources and destinations, both sets of nodes are updated at once, similar to the `update_at_start` strategy
Could you give some guidance on how to train the model on multiple GPUs? I have a very large dataset and I run out of memory (32 GB total). Should...
i have dependency issues while importing paddle. ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the...
