💾 Archived View for dioskouroi.xyz › thread › 29379936 captured on 2021-11-30 at 20:18:30. Gemini links have been rewritten to link to archived content
-=-=-=-=-=-=-
________________________________________________________________________________
At work I've had to create a model to rank our products and was searching for implementations and working examples of Learn To Tank models (especially with lightGBM examples). There are actually very few tutorials out there. I understand the theory but there are a thousand questions I had about the implementation and the data preparation and could not find the answer on the internet.
It was quite surprised given how much Machine Learning has been blog spammed.
I recently completed a Learn To Rank (with LightGBM) recommendation system at work and I agree. For how powerful the libraries are you would think there would be many more examples. Have you collected or posted your questions anywhere? I hit an error I didn't understand and posted questions on a related LightGBM issue and LightGBM devs responded with enough detail to at least get me unstuck.
I work in the similar space (recsys). Learning-To-Rank sounds amazing on paper but logging "good" (yes, those quotes are warranted) ranked data in production is damn near impossible. Every time you come up with strategy X, someone points out how its statistically biased to hell.