MultiStepStateScheduler#
- class ignite.handlers.state_param_scheduler.MultiStepStateScheduler(initial_value, gamma, milestones, param_name, save_history=False, create_new=False)[source]#
Update a parameter during training by using a multi step function. The function decays the parameter value by gamma once the number of steps reaches one of the milestones. Based on MultiStepLR from PyTorch. https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.MultiStepLR.html
- Parameters
initial_value (float) – Starting value of the parameter.
gamma (float) – Multiplicative factor of parameter value decay.
milestones (List[int]) – List of step indices. Must be increasing.
param_name (str) – name of parameter to update.
save_history (bool) – whether to log the parameter values to engine.state.param_history, (default=False).
create_new (bool) – whether to create
param_nameonengine.statetaking into account whetherparam_nameattribute already exists or not. Overrides existing attribute by default, (default=False).
Examples
param_scheduler = MultiStepStateScheduler( param_name="param", initial_value=1, gamma=0.9, milestones=[3, 6, 9, 12] ) # parameter is param, initial_value sets param to 1, gamma is set as 0.9 # Epoch 1 to 2, param does not change as milestone is 3 # Epoch 3, param changes from 1 to 1*0.9, param = 0.9 # Epoch 3 to 5, param does not change as milestone is 6 # Epoch 6, param changes from 0.9 to 0.9*0.9, param = 0.81 # Epoch 6 to 8, param does not change as milestone is 9 # Epoch 9, param changes from 0.81 to 0.81*0.9, param = 0.729 # Epoch 9 to 11, param does not change as milestone is 12 # Epoch 12, param changes from 0.729 to 0.729*0.9, param = 0.6561 param_scheduler.attach(default_trainer, Events.EPOCH_COMPLETED) @default_trainer.on(Events.EPOCH_COMPLETED) def print_param(): print(default_trainer.state.param) default_trainer.run([0], max_epochs=12)
1.0 1.0 0.9 0.9 0.9 0.81 0.81 0.81 0.7290... 0.7290... 0.7290... 0.6561
New in version 0.4.7.
Methods
Method to get current parameter values