airalcorn2 / batter-pitcher-2vec

A model for learning distributed representations of MLB players.

batter-pitcher-2vec's Issues

Sigmoid activation for embedding layers

Hey Michael,

I enjoyed your presentation of this project at Sloan this year, and it's great that you posted the code up here, so thanks for that! I was revisiting it recently and I was wondering if there was any specific reason you choice sigmoid activation functions for the batter/pitcher embedding layers. Is this a typical feature of embedding layers? The only reason I could think of outside of the optimization of the network itself would be to have these embeddings be on a uniform scale for the sake of comparison and analysis after these parameters had been learned. I'm a relative novice when it comes to Neural Networks so any insight here would be greatly appreciated.

Thanks,
Johannes