Basic Weight and Biases with Code

Weight and Biases

• Easy to share the experiemnt with other people!
• can track hardware status
• install with pip

  !pip install wandb -q
  

• make config file

  config={"epochs":EPOCHS, "batch_size":BATCH_SIZE, "learning_rate":LEARNING_RATE}
  # initialize. set entity name from web
  wandb.init(project="project_name", entity="entity_name")
  # after setting the configuartion dictionary
  wandb.init(project="project_name", config=config)
  # setting configuration 
  wandb.config.batch_size = BATCH_SIZE
  wandb.config.learning_rate = LEARNING_RATE
  

• how to use in the model

  for e in range(1, EPOCHS+1):
    epoch_loss = 0
    epoch_acc = 0
    for X_batch, y_batch in train_dataset:
        X_batch, y_batch = X_batch.to(device), y_batch.to(device).type(torch.cuda.FloatTensor)
        ...
  
        ...
    # saving the log
    wandb.log({'accuary':train_acc, 'loss':train_loss})