Hi Jeroen, thanks for getting in touch!
This is an interesting area. AirBNB did a really nice writeup of turning browsing paths into an embedding (which then drives recommendations) and you could use purchase paths in the same way. Check this out:
There are a lot of approaches you can take, one of the fun things about graph-ML is the number of different solutions. You could try:
I totally agree that this is very simplistic; We often get asked about how to connect different machine learning libraries with Neo4j and wanted to get out some preliminary code for Keras.
If you check out our other articles, you’ll hopefully find them more intellectually fulfilling :)
Yes. We eventually shifted to having read slaves and write masters, and moving the problematic queries (often from BI) to isolated slaves. We also pumped up the machine sizes. The biggest pain was the lack of clear monitoring and configuration tools in RDS, it gave us little indication what was going on under the hood.