ML-Purdue
/
aqacf-ml-main


  
1

	
2

	
3

	
4

	
5

	
6

	
7

	
8

	
9

	
10

	
11

	
12

	
            1. in main,  adam optimizer, the weight decay should be on for stage 2, off for stage 1
	optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
   	optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate, weight_decay = 1e-4)
2. in main, the scheduler should be StepLR, then CosineAnnealingLR
	 scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size = epochs//2, gamma = 0.1)
	scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=epochs, eta_min=0)
3. in main, the checkpoint path should be nothing for stage 1, the checkpoint for stage 2
	checkpoint_path = ''
	checkpoint_path = r"C:\Users\Michael\aqacf-ml-main-new\aqacf-ml-main\model_output_111302242025\state_epoch_00010.pt"
4. in training_config, set the first stage to 20 epochs and the second to 10 epochs
5. in main, the dropout should be off for stage 1, on for stage 2