Zecheng Zhang

Hi, Yes, the current `remove_self_loops=True` will be problematic when the node types are different. Thanks for pointing out this problem! I will update the `HeteroSAGEConv` layer recently.

> thanks @zechengz . i see, looks like `snapx` comes as a part of `snap` which is the pypi package `snap-stanford`? > > what do you think about this: >...

Does PyTorch Geometric successfully be installed?

That's a little bit strange. Because DeepSNAP is almost based on PyTorch, PyG and NetworkX, and all other implementations are just pure Python. The kind of error like `Illegal instruction...

Hi, Thanks for pointing out this. Right, handling the graph data by using the NetworkX graph object seems not efficient. But the performance issue for generating the `HeteroGraph` might be...

Hi, The idea of using separate linear layers for self and neighbor is mainly derived from the Relational GCN, which is briefly described in P13 of this [slides](https://web.stanford.edu/class/cs224w/slides/10-kg.pdf). And adding...

Hi, sorry about the late reply. Basically, `EnsembleGenerator` is a generator composed of a list of generators. Those generators will have `gen_prob` probabilities to be sampled or used to `generate`...

Thanks @SebastianHurubaru ! Sorry that I am very busy recently. If possible, can you also create a PR for the fix?

Why we need this?

@Wendong-Fan In the agent app how we setup the chat agent? For example, can we set the OpenAI Key via environment variable at somewhere instead of passing the api key...