config-defaults.yaml

model_type:
  desc: The backbone network of meta model, it can be ConvNet4, ConvNet6, ResNet12, ResNet18, ResNet34, WRN28
  value: "ConvNet4"
dataset:
  desc: The dataset can be CUB, MiniImageNet, TieredImageNet, CIFAR100
  value: "CUB"

num_epochs:
  desc: Number of epochs to train over
  value: 2000
save_best:
  desc: Save the best model
  value: True
if_augmentation:
  desc: Use augmentation methods
  value: True
batch_size:
  desc: Batch size
  value: 1
sgd_lr:
  desc: Learning rate of SGD
  value: 0.1
patience:
  desc: patience of optimizer scheduler
  value: 50
reduce_factor:
  desc: reduce rate of optimizer sechduler
  value: 0.5

num_way:
  desc: number of way
  value: 5
num_shot:
  desc: number of shot
  value: 1
num_query:
  desc: The number of query samples in a task
  value: 15
num_train:
  desc: The number of train tasks in an epoch
  value: 100
num_val:
  desc: The number of validation tasks
  value: 600
num_test:
   desc: The number of test tasks
   value: 600

proxy_type:
   desc: Proxy type can be Proxy, Mean, Sum
   value: "Mean"
classifier:
   desc: Classifier can be 3DConv, Euclidean, FC
   value: "Euclidean"
gpu_id:
   desc: gpu id
   value: "0"

teacher_model:
   desc: teacher model, Glove or FastText or Bert
   value: "Bert"
random_embedding:
   desc: embed the labels randomly
   value: False
temperature:
   desc: temperature used in attentnion transfer
   value: 10
lam:
   desc: lamda of text modality loss
   value: 0.3
optimizer:
   desc: SGD or Adam or AdamW
   value: "Adam"
aux_loss:
   desc: MSE or KL
   value: "KL"
scale:
   desc: Scale value in cross attention
   value: 0.4