You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
For probing tasks, we used an MLP with a Sigmoid nonlinearity and and tuned the nhid (in [50, 100, 200]) and dropout (in [0.0, 0.1, 0.2]) on the dev set.
However, in the code it looks like the parameters given by the user are always used. No tuning takes place and no predefined hyperparameters are loaded. Maybe I missed something?
Should I do hyperparameter tuning to get results that are comparable to the literature?
The text was updated successfully, but these errors were encountered:
The readme states
For probing tasks, we used an MLP with a Sigmoid nonlinearity and and tuned the nhid (in [50, 100, 200]) and dropout (in [0.0, 0.1, 0.2]) on the dev set.
However, in the code it looks like the parameters given by the user are always used. No tuning takes place and no predefined hyperparameters are loaded. Maybe I missed something?
Should I do hyperparameter tuning to get results that are comparable to the literature?
The text was updated successfully, but these errors were encountered: