• I've trained a few models on the embeddings; each of these train files include evaluation metrics.
• They've been tested on samples from the CB data (I realise the file paths are to my local machine, so they'd need to be changed to retrieve the data from wherever you get your data. Probably would have been better to retrieve from S3 to make the code more reproducible, so a good learning ha).
• The test_cb_logreg notebook has a couple of options for choose your sample - either with 100 random records or those which contain "heat pump".

Hopefully the code is a useful start for you to pick up 🙂