Pretrain for both the classifier and adversarial model. You can select which model it is by setting model_type parameter (it customizes the prints properly). Pretrain can also collect more data to training analysis and print it out which is useful for monitoring the learning process and making adjustments.
pretrain_net( n_epochs = 15, model, dsl, model_type, learning_rate = 0.001, sensitive_test, dev, verbose = TRUE, monitor = TRUE )
| n_epochs | integer setting number of epochs for training. Default: 15 |
|---|---|
| model | net, nn_module, neural network model we want to train |
| dsl | dataset_loader object for the training |
| model_type | indicates which model we train (0 for preadversarial, 1 for preclassifier, 2 for classifier only) |
| learning_rate | float from [0,1] setting learning rate for model. Default: 0.001 |
| sensitive_test | test vector for sensitive variable used to calculate STP |
| dev | device used to calculations (cpu or gpu) |
| verbose | logical indicating if we want to print monitored outputs or not |
| monitor | logical indicating if we want to monitor the learning process or not (monitoring tends to slow down the training process, but provides some useful info to adjust parameters and training process) |
list(train_loss,test_loss,optimizer)
if (FALSE) { dev <- "cpu" # presaved output of preprocess function processed <- torch_load(system.file("extdata","processed",package="fairpan")) dsl <- dataset_loader(processed$train_x, processed$train_y, processed$test_x, processed$test_y, batch_size=5, dev=dev) # presaved torch model model <- torch_load(system.file("extdata","clf1",package="fairpan")) pretrain_net( n_epochs = 1, model = model, dsl = dsl, model_type = 1, learning_rate = 0.001, sensitive_test = processed$sensitive_test, dev=dev, verbose = TRUE, monitor = TRUE ) }