FedPer - Set split point of neural network at different locations about fl-bench

[KarhouTam]

Hi, @WilliamLindskog.

You can customize the self.classifier in model class.

FL-bench/src/config/models.py

Lines 262 to 266 in a007582

	self.base = models.resnet18(
	weights=models.ResNet18_Weights.DEFAULT if pretrained else None
	)
	self.classifier = nn.Linear(self.base.fc.in_features, config[dataset])
	self.base.fc = nn.Identity()

but please ensure that the self.base and self.classifier contain all modules in your model.

[WilliamLindskog]

Hi, @KarhouTam,

Thank you for your prompt answer. With regards to ResNet, if I were to put the split point somewhere else rather than the final layer, would the following functions forward, get_all_features, and get_final_features generalize or would I need to customize them as well?

[KarhouTam]

You do not need to change these methods. These methods are not called by FedPer, except forward(),which should work properly if you set the base and classifier well.