Edge properties for reasonable edge classification 📊

[KristinaUlicna]

PR contribution summary

Why is this PR useful / good for? Please describe the problem(s) you're trying to address.

• Computes, stores & organises a set of geometrical EdgeProperties
• Implements edge properties into the classifier training
• Extends available model catalogue by those which input edge properties
• Implements the architecture for Graph Attention Network (GAT)
• Allows [multi-head] attention visualisation to monitor training progress
• Adds additional config hyperparameters to keep up with the progress

List of proposed changes / linked issues & discussions

• ✅ Resolves [DEVELOPMENT] Compute edge properties #303
• ✅ Resolves [DEVELOPMENT] Define an EdgePropertyCruncher #304
• ✅ Resolves Add more sophisticated classifier model options #291 ▶️ includes GAT model which computes edge attention weights based on edge properties
• ✅ Resolves [DEVELOPMENT] Store edge properties in Data object for training #284 ▶️ datasets.py
• ✅ Resolves [DEVELOPMENT] Allow specifying variable number of graph conv & dense linear layers #281 ▶️ classifier.py
• ✅ Resolves [FEATURE REQUEST] Consider visualising the edge properties in the dataset  #294 ▶️ via edge_properties.ipynb
• ✅ Fixes [DEVELOPMENT] How to set a global config with different font attributes in matplotlib? #247 with seaborn import from styling in relevant __init__.py
• ✅ Closes [DEVELOPMENT] Utilise edge_attr in graph convolutions #220
• ✅ Closes [DEVELOPMENT] Use distance / weighting as feature for edge embedding learning #278

What should a reviewer concentrate their feedback on?

• 🏃 Scripts to run - reproducing training results
• 💻 Code quality

What type of PR is this? (check all applicable)

• 🪄 Feature
• #️⃣ Documentation / code annotation
• 🧑‍💻 Code refactor / style
• 🔥 Performance Improvements

Added tests?

• 🙅 no, temporarily silenced some due to lack of graph attributes

---

PR review summary

Describe what this PR does & how you reviewed the individual items, where needed:

Some helper checks to tick off:

• Focus on image annotation
• Focus on model training
• Could any optimization be applied?
• Is there any redundant code?
• Are there any spelling errors?

In conclusion, after my review, I'd like to:

• 🙋 ask some clarifying questions
• 🙅 suggest some specific changes

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[KristinaUlicna]

Temporary silencing some tests which keep failing due to lack of edge properties attribute in the sample graphs. Fix upcoming in subsequent PR(s) 🚀

[crangelsmith]

I think you can skip tests without having to comment them out with some of these options https://docs.pytest.org/en/7.3.x/how-to/skipping.html

[KristinaUlicna]

Thanks for the suggestion - this is gold 🥇 ! All failing tests are now re-implemented in PR #308 (which extends the failing run if no edge properties are created)

[crangelsmith]

Ran the training (after calculating node and edge properties beforehand), I get the following results:

I think this is compatible with @KristinaUlicna results :)

[crangelsmith]

My main suggestions (mostly discussed offline with @KristinaUlicna) are being implemented in the following PRs, so happy to approve this one.