Additional structured Rossmann data

Hi there, I was wondering what would need to change about the features presented to the neural network if more data became available - let’s say that we now knew:

  • the products that were sold in each store (with all the usual related data about products like color, type of product, etc)
  • the components that make up a product (things like LEDs, actuators, servos, plastic shells, screws - I know this is a little far fetched but let’s just say for the sake of argument that we knew this data)

Is this additional data something that’s appropriate to be represented in an embedding? Or, since we have columnar oriented data arranged in rows would the additional data be output in rows that have partially duplicated data?

Thanks, just trying to understand how to feed data to the NN that’s spread out in relational tables!

With more data, it would help to find columns that could help join them together with the other csv records. And yes, they’d probably be categorical variables themselves, and therefore representable as embeddings.

One point I’ve been slightly confused on - are embeddings really meant for time series type data or can it also be used for data related to the entity at hand (eg. data about items sold at a particular store)?