search

bio-ontology-research-group / deepgo

Function prediction using a deep ontology-aware classifier
http://deepgo.bio2vec.net
76 stars 24 forks source link

Learning rate for training #17

Closed datduong closed 4 years ago

datduong commented 5 years ago

Hi, would you be able to tell me the learning rate and batch size used for each MF, CC, BP ontology? I see that the default is 0.001 in the model that uses only amino sequence. Is 0.001 used for all experiments? Thanks.

coolmaksat commented 5 years ago

I tried different learning rates ranging from 0.0001 to 0.1, but If I'm not mistaken 0.001 was used in the latest experiments.

datduong commented 5 years ago

Do you know if Keras optimizer has some special initialization for unknown parameters? I implemented your code in Pytorch. I am getting very similar AUC. However Fmax value is consistently lower than Keras (about 4%) lower.

Thanks.

coolmaksat commented 5 years ago

I'm not sure. You can compare the default parameters https://keras.io/optimizers/