pytorch_geometric_temporal VS pytorch_geometric

I have done this a few times now. Alone (e.g. https://github.com/paulmorio/geo2dr) and in collaboration with others (e.g. https://github.com/benedekrozemberczki/pytorch_geometric_tem...) primarily as a way to learn about the methods I was interested in from a research perspective whilst improving my skills in software engineering. I am still learning.

Starting out I would recommend implementing fundamental building blocks within whatever 'subculture' of ML you are interested in whether that be DL, kernel methods, probabilistic models, etc.

Let's say you are interested in deep learning methods (as that's something I could at least speak more confidently about). In that case build yourself an MLP layer, then an RNN layer, then a GNN layer, then a CNN layer, and an attention layer along with some full models with those layers on some case studies exhibiting different data modalities (images, graphs, signals). This should give you a feel for the assumptions driving the inductive biases in each layer and what motivates their existence (vs. an MLP). It also gives you the all the building blocks you can then extend to build every other DL layer+model out there. Another reason is that these fundamental building blocks have been implemented many times so you have a reference to look to when you get stuck.

On that note: here are some fun GNN papers to implement in order of increasing difficulty (try building using vanilla PyTorch/Jax instead of PyG).

Repo: https://github.com/benedekrozemberczki/pytorch_geometric_temporal

https://github.com/pyg-team/pytorch_geometric/tree/master/examples any of the scripts ending in _segmentation.py can be used for semantic segmentation of point clouds

Here the code I'm referring to: https://github.com/pyg-team/pytorch_geometric/blob/master/examples/hetero/metapath2vec.py

Today version 2.3 got released: https://github.com/pyg-team/pytorch_geometric/releases/tag/2.3.0

Here is a working example for semantic segmentation with pointnet++ using PyTorch geometric. There are equivalent scripts for dgcnn, randlanet, point transformer in the same folder https://github.com/pyg-team/pytorch_geometric/blob/master/examples/pointnet2_segmentation.py

I don't know if any RL library includes an already implemented agent that can process graphs. However there are a number of deep learning frameworks that can help with the implementation of graph neural networks, especially Graph Nets (based on Tensorflow) and PyTorch Geometric. You might need to modify an existing RL agent to make use of one of these frameworks. If you are not familiar with GNNs you can look up these surveys. This article may also be of interest to you: it tackles graph-based environments, and the paper's code is available (it has a custom implementation of A2C and uses PyTorch Geometric -- btw it doesn't use Gym's space.graph since this feature is very recent in Gym).

Meanwhile, PyTorch-Geometric is 3 years old and 13K stars on Github.

https://github.com/pyg-team/pytorch_geometric

A pytorch implementation: https://github.com/rusty1s/pytorch_geometric