Parameters

Note

Your parameters in your configuration file must match the keywords in __init__ of your model, optimizer, learning rate scheduler, or loss criterion. For example, if you want to set base_lr in the learning scheduler CyclicLR, you will need:

[parameters]
[parameters.learning_rate]
base_lr = 0.01

model

All the parameter settings that need to be passed as keyword parameters when initializing the model, such as num_classes or cut_layer. The set of parameters permitted or needed depends on the model.

optimizer

All the parameter settings that need to be passed as keyword parameters when initializing the optimizer, such as lr, momentum, or weight_decay. The set of parameters permitted or needed depends on the optimizer.

learning_rate

All the parameter settings that need to be passed as keyword parameters when initializing the learning rate scheduler, such as gamma. The set of parameters permitted or needed depends on the learning rate scheduler.

loss_criterion

All the parameter settings that need to be passed as keyword parameters when initializing the loss criterion, such as size_average. The set of parameters permitted or needed depends on the loss criterion.