class

LambdaLR

extends_LRScheduler

LambdaLR(optimizer: Optimizer, lr_lambda: Callable[[int], float] | list[Callable[[int], float]], last_epoch: int = -1, verbose: bool = False)

source

Set the learning rate using a user-supplied multiplicative factor function.

At each epoch $t$ , the learning rate for param group $i$ is:

\eta_t^{(i)} = \eta_0^{(i)} \cdot \lambda_i(t)

where $\eta_0^{(i)}$ is the initial learning rate of group $i$ and $\lambda_i$ is the corresponding callable.

Parameters

optimizerOptimizer

Wrapped optimizer.

lr_lambdacallable or list of callable

A function f(epoch: int) -> float returning the multiplicative factor for the learning rate. If a single callable is given it is applied to every param group; if a list is given it must have the same length as optimizer.param_groups.

last_epochint= -1

The index of the last epoch (default: -1).

verbosebool= False

Print the updated LR after each step if True (default: False).

Attributes

lr_lambdaslist of callable

One factor function per param group.

Notes

LambdaLR provides maximum flexibility: any monotone or cyclic schedule can be encoded as a Python function. For a linearly increasing warmup followed by constant LR:

.. code-block:: python

warmup = 5
fn = lambda t: min(1.0, t / warmup)
scheduler = optim.LambdaLR(optimizer, lr_lambda=fn)

Examples

>>> import lucid.optim as optim
>>> optimizer = optim.Adam(model.parameters(), lr=1e-3)
>>> scheduler = optim.LambdaLR(optimizer, lr_lambda=lambda t: 0.95 ** t)
>>> for epoch in range(50):
...     train(...)
...     optimizer.step()
...     scheduler.step()

Methods (2)

dunder

init

→None

__init__(optimizer: Optimizer, lr_lambda: Callable[[int], float] | list[Callable[[int], float]], last_epoch: int = -1, verbose: bool = False)

source

Initialise the LambdaLR. See the class docstring for parameter semantics.

get_lr

→list[float]

get_lr()

source

Compute the learning rate for each parameter group at the current step.

Returns

list[float]

One learning rate per param group, derived from the schedule formula documented in the class docstring.