Skip to content

Schedulers

simplegrad.schedulers.func_based.LinearLR

Bases: Scheduler

Source code in simplegrad/schedulers/func_based.py 
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
class LinearLR(Scheduler):
    def __init__(
        self,
        optimizer: Optimizer,
        start_lr: float | None,
        end_lr: float | None,
        total_steps: int | None,
        rate: float | None,
    ) -> None:
        """
        Possible combinations of parameters:
        1. start_lr, end_lr, total_steps
        2. start_lr, end_lr, rate
        3. start_lr, total_steps, rate
        4. end_lr, total_steps, rate
        5. start_lr, rate (assumes infinite total steps)
        """
        super().__init__(optimizer)
        self.start_lr = start_lr
        self.end_lr = end_lr
        self.total_steps = total_steps
        self.rate = rate

        if (
            start_lr is not None
            and end_lr is not None
            and total_steps is not None
            and rate is not None
        ):
            raise ValueError(
                "Parameter mismatch. Only three of the parameters start_lr, end_lr, total_steps, rate should be provided (or two: start_lr, rate)."
            )

        # Case 1
        if start_lr is not None and end_lr is not None and total_steps is not None:
            self.rate = (end_lr - start_lr) / total_steps
        # Case 2
        elif start_lr is not None and end_lr is not None and self.rate is not None:
            self.total_steps = int((end_lr - start_lr) / self.rate)
        # Case 3
        elif start_lr is not None and total_steps is not None and self.rate is not None:
            self.end_lr = start_lr + total_steps * self.rate
        # Case 4
        elif end_lr is not None and total_steps is not None and self.rate is not None:
            self.start_lr = end_lr - total_steps * self.rate
        # Case 5
        elif start_lr is not None and self.rate is not None:
            self.total_steps = float("inf")

    def step(self, *args, **kwargs) -> None:
        if self.steps < self.total_steps:
            new_lr = self.start_lr + self.rate * self.steps
            self.optimizer.set_lr(new_lr)
        self.steps += 1

__init__(optimizer: Optimizer, start_lr: float | None, end_lr: float | None, total_steps: int | None, rate: float | None) -> None

Possible combinations of parameters: 1. start_lr, end_lr, total_steps 2. start_lr, end_lr, rate 3. start_lr, total_steps, rate 4. end_lr, total_steps, rate 5. start_lr, rate (assumes infinite total steps)

Source code in simplegrad/schedulers/func_based.py 
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
def __init__(
    self,
    optimizer: Optimizer,
    start_lr: float | None,
    end_lr: float | None,
    total_steps: int | None,
    rate: float | None,
) -> None:
    """
    Possible combinations of parameters:
    1. start_lr, end_lr, total_steps
    2. start_lr, end_lr, rate
    3. start_lr, total_steps, rate
    4. end_lr, total_steps, rate
    5. start_lr, rate (assumes infinite total steps)
    """
    super().__init__(optimizer)
    self.start_lr = start_lr
    self.end_lr = end_lr
    self.total_steps = total_steps
    self.rate = rate

    if (
        start_lr is not None
        and end_lr is not None
        and total_steps is not None
        and rate is not None
    ):
        raise ValueError(
            "Parameter mismatch. Only three of the parameters start_lr, end_lr, total_steps, rate should be provided (or two: start_lr, rate)."
        )

    # Case 1
    if start_lr is not None and end_lr is not None and total_steps is not None:
        self.rate = (end_lr - start_lr) / total_steps
    # Case 2
    elif start_lr is not None and end_lr is not None and self.rate is not None:
        self.total_steps = int((end_lr - start_lr) / self.rate)
    # Case 3
    elif start_lr is not None and total_steps is not None and self.rate is not None:
        self.end_lr = start_lr + total_steps * self.rate
    # Case 4
    elif end_lr is not None and total_steps is not None and self.rate is not None:
        self.start_lr = end_lr - total_steps * self.rate
    # Case 5
    elif start_lr is not None and self.rate is not None:
        self.total_steps = float("inf")

simplegrad.schedulers.func_based.ExponentialLR

Bases: Scheduler

Source code in simplegrad/schedulers/func_based.py 
60
61
class ExponentialLR(Scheduler):
    pass

simplegrad.schedulers.func_based.CosineAnnealingLR

Bases: Scheduler

Source code in simplegrad/schedulers/func_based.py 
64
65
class CosineAnnealingLR(Scheduler):
    pass

simplegrad.schedulers.metric_based.ReduceLROnPlateauLR

Bases: Scheduler

Source code in simplegrad/schedulers/metric_based.py 
4
5
class ReduceLROnPlateauLR(Scheduler):
    pass