search

edgarschnfld / CADA-VAE-PyTorch

Official implementation of the paper "Generalized Zero- and Few-Shot Learning via Aligned Variational Autoencoders" (CVPR 2019)
MIT License
283 stars 57 forks source link

questions about the hyperparameters #23

Closed KernLC closed 4 years ago

KernLC commented 4 years ago

Hi, I don't understand what the cls_train_steps and num_shots parameters mean. Could you please explain them for me simply?

edgarschnfld commented 4 years ago

Hi, Training proceeds in two phases: First we train the VAE to learn the shared embedding, then we train the classifier separately on embedded features. cls_train_steps is the number of training steps of that classifier. Classifier training relies on early stopping, so the cls_trainsteps were determined on the hyperparameter tuning split to know when to end training. num_shots is the number of shots: if it is 0 we do 0-shot learning, if it is 1 we do 1-shot learning, and so on.

KernLC commented 4 years ago

Hi, Training proceeds in two phases: First we train the VAE to learn the shared embedding, then we train the classifier separately on embedded features. cls_train_steps is the number of training steps of that classifier. Classifier training relies on early stopping, so the cls_trainsteps were determined on the hyperparameter tuning split to know when to end training. num_shots is the number of shots: if it is 0 we do 0-shot learning, if it is 1 we do 1-shot learning, and so on.

Thanks a lot!!!!