Skip to content

configs

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

configs

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
SL
SL
 
 
SSL
SSL
 
 
yaml
yaml
 
 
README.md
README.md
 
 

README.md

The following are the parameters that can be easily configured by the users:

  • setting: Type of learning method
  • dataset
    • name: Name of the dataset
    • datadir: Directory where the dataset exists/ to download
    • feature: dss/classimb
    • type: pre-defined
  • dataloader
    • shuffle: Reshuffle the data during every epoch
    • batch_size: Number of samples per batch
    • pin_memory: To transfer fetched data to CUDA enabled GPUs
  • model
    • architecture: Network architecture used for training
    • type: pre-defined
    • num_classes: Number of target classes in the dataset
  • ckpt
    • is_load: To load previously saved checkpoint
    • is_save: To save checkpoints
    • dir: Directory to save the model checkpoints
    • save_every: Save every N epochs
  • loss
    • type: loss function
    • use_sigmoid: True/False
  • optimizer
    • type: Type of optimizer
    • momentum: Momentum factor
    • lr: Learning rate
    • weight_decay: Weight decay
  • scheduler
    • type: Type of scheduler
    • T_max: Maximum number of iterations
  • dss_strategy
    • type: Subset Selection Algorithm
    • fraction: Percentage of data used for training
    • select_every: Subset selection every N epochs
    • kappa: Kappa value
    • lam: Lambda value
    • valid: For class imbalance training
  • train_args
    • num_epochs: Number of epochs to train the model
    • device: Device used for training - CPU/GPU
    • print_every: Print every N epochs
    • results_dir: Output log directory
    • print_args: Values to be logged to file - val_loss, val_acc, tst_loss, tst_acc, trn_loss, trn_acc, subtrn_loss, subtrn_acc, time
    • return_args: Arguments to be returned