Opened this issue a year ago · 1 comments
Adding a unique (learned) vector for each IID to all the embedding vector lead to a perfect fit (thankfully):
It is indeed possible to overfit if we add in IID info to the data