Y
Hacker News
new
|
ask
|
show
|
jobs
by
moultano
3326 days ago
Much better to include the feature in the training, and exclude it from the inference. That actually cancels out the effect, rather than just incentivizing the model to reconstruct it.