search

automl / Auto-PyTorch

Automatic architecture search and hyperparameter optimization for PyTorch
Apache License 2.0
2.38k stars 289 forks source link

Dropout application #188

Open ArlindKadra opened 3 years ago

ArlindKadra commented 3 years ago

I have a question regarding dropout and how it is applied. In the previous non-refactored version of AutoPyTorch there was a dropout_shape hyperparameter, which does not seem to be there anymore. Is now dropout applied with the same rate at every layer and if not, does it follow the network shape then ?

ravinkohli commented 3 years ago

Hey, as far as I can tell, dropout_shape was not a hyperparameter but a parameter that depended on the network shape in the master branch. Similarly, it has exactly the same application in 'refactor_development'. You can find it here and for reference, it is implemented using this

ravinkohli commented 2 years ago

This will also be fixed in #358