StepStateScheduler

class ignite.handlers.state_param_scheduler.StepStateScheduler(initial_value, gamma, step_size, param_name, save_history=False, create_new=False)

Update a parameter during training by using a step function. This function decays the parameter value by gamma every step_size. Based on StepLR from PyTorch. https://pytorch.org/docs/stable/generated/torch.optim.lr_scheduler.StepLR.html

Parameters

• initial_value (float) – Starting value of the parameter.

• gamma (float) – Multiplicative factor of parameter value decay.

• step_size (int) – Period of parameter value decay.

• 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_name on engine.state taking into account whether param_name attribute already exists or not. Overrides existing attribute by default, (default=False).

Examples

from collections import OrderedDict

import torch
from torch import nn, optim

from ignite.engine import *
from ignite.handlers import *
from ignite.metrics import *
from ignite.utils import *
from ignite.contrib.metrics.regression import *
from ignite.contrib.metrics import *

# create default evaluator for doctests

def eval_step(engine, batch):
    return batch

default_evaluator = Engine(eval_step)

# create default optimizer for doctests

param_tensor = torch.zeros([1], requires_grad=True)
default_optimizer = torch.optim.SGD([param_tensor], lr=0.1)

# create default trainer for doctests
# as handlers could be attached to the trainer,
# each test must define his own trainer using `.. testsetup:`

def get_default_trainer():

    def train_step(engine, batch):
        return batch

    return Engine(train_step)

# create default model for doctests

default_model = nn.Sequential(OrderedDict([
    ('base', nn.Linear(4, 2)),
    ('fc', nn.Linear(2, 1))
]))

manual_seed(666)

default_trainer = get_default_trainer()

param_scheduler = StepStateScheduler(
    param_name="param", initial_value=1, gamma=0.9, step_size=5, create_new=True
)

# parameter is param, initial_value sets param to 1, gamma is set as 0.9
# Epoch 1 to 4, param does not change as step size is 5,
# Epoch 5, param changes from 1 to 1*0.9, param = 0.9
# Epoch 5 to 9, param = 0.9 as step size is 5,
# Epoch 10, param changes from 0.9 to 0.9*0.9, param = 0.81
# Epoch 10 to 14, param = 0.81, as step size is 5
# Epoch 15, param changes from 0.81 to 0.81*0.9, param = 0.729
# and so on ... the param change at Epoch = 5, 10, 15, 20, . . .

param_scheduler.attach(default_trainer, Events.EPOCH_COMPLETED)

@default_trainer.on(Events.EPOCH_COMPLETED(every=5))
def print_param():
    print(default_trainer.state.param)

default_trainer.run([0], max_epochs=25)

0.9
0.81
0.7290...
0.6561
0.5904...

New in version 0.4.7.

Methods

get_param

Method to get current parameter values

get_param()

Method to get current parameter values

Returns

list of params, or scalar param

Return type

Union[List[float], float]