according to fastText's autotuning blog entry "(the) strategy to explore various hyperparameters is inspired by existing tools, such as Nevergrad, but tailored to fastText by leveraging the specific structure of models". I am trying to follow and understand the autotuning source code but can't figure out what algorithm or strategy is being followed. I used to think it was a kind of Nevergrad's differential evolution with gaussian initialization. I think there is no available documentation about the details. Can you shed some light about it?
MrRace
commented
3 years ago
Hi,
according to fastText's autotuning blog entry "(the) strategy to explore various hyperparameters is inspired by existing tools, such as Nevergrad, but tailored to fastText by leveraging the specific structure of models". I am trying to follow and understand the autotuning source code but can't figure out what algorithm or strategy is being followed. I used to think it was a kind of Nevergrad's differential evolution with gaussian initialization. I think there is no available documentation about the details. Can you shed some light about it?
Thanks!