Arguments

Arguments supported in the kclassify model as hyperparameters.

Argument	Description	Default
--base_model	vgg16, vgg19, efficientnetB0, mobilenetv2	efficientnetB0
--batch_size	(optional) Batch size	32
--lr	(optional) Learning rate	.01
--has_wandb	(optional) Logs to the wandb server	False
--preprocessor	use the model preprocessor on the inputs	False
--featurewise_normalize	use featurewise centering and std normalizing	True
--train_stats	(optional) configuration file with training image statistics; must exist when using the --featurewise_normalize option	-
--epochs	(optional) Number of epochs to train	1
--optimizer	(optional) adam, radam, ranger	adam
--loss	(optional) Type of loss function for the gradients: for the gradients categorical_crossentropy, or categorical_focal_loss	categorical_crossentropy
--dropout	(optional) Add a drop out layer	False
--horizontal_flip	(optional) Add horizontal flip augmentation during training	False
--vertical_flip	(optional) Add vertical flip augmentation during training	False
--early_stop	(optional) Add early stopping	False
--rotation_range	(optional) Apply rotation augmentation between 0-1 as percent of image size during training	0.0
--augment_range	(optional) Apply width, shift, and zoom augmentation during training 0-1 as percent of image size	0.0
--shear_range	(optional) Apply sheer augmentation during training 0-1 as percent of image size	0.0