an end to end pipeline (from data ingestion through training to evaluation) is now available!

[radekosmulski]

I just wanted to give a quick update that I just pushed a new notebook to the repository. In this notebook we

• load the data we processed in the first notebook
• define an auto encoder architecture
• train the model
• evaluate results and compare with the embeddings shared by the authors of the paper we are working on implementing

The model trains but unfortunately, in the minimal testing that I've done, it doesn't seem to be learning a well structured embedding space. There are many tweaks that could be done to the model, it is also not unlikely that we have a bug somewhere.

I am in the process of creating and testing out a mini dataset to experiment not. In the meanwhile I am also planning to spend a bit more time going over the architecture and data processing and will reach out to one of the authors of the paper asking for guidance. Will report back here as I make progress on this.

[bs]

@radekosmulski, is there anything more we need from this cycle of work before closing this?

[radekosmulski]

don't believe so - closing