otx.core.optimizer#
Modules related to an optimizer.
Classes
|
Optimizer callable supports OTX hyper-parameter optimization (HPO) algorithm. |
- class otx.core.optimizer.OptimizerCallableSupportHPO(optimizer_cls: type[Optimizer] | str, optimizer_kwargs: dict[str, int | float | bool])[source]#
Bases:
objectOptimizer callable supports OTX hyper-parameter optimization (HPO) algorithm.
It makes OptimizerCallable pickelable and accessible to parameters. It is used for HPO and adaptive batch size.
- Parameters:
optimizer_cls – Optimizer class type or string class import path. See examples for details.
optimizer_kwargs – Keyword arguments used for the initialization of the given optimizer_cls.
Examples
This is an example to create MobileNetV3ForMulticlassCls with a SGD optimizer and custom configurations.
```python from torch.optim import SGD from otx.algo.classification.mobilenet_v3_large import MobileNetV3ForMulticlassCls
- model = MobileNetV3ForMulticlassCls(
num_classes=3, optimizer=OptimizerCallableSupportHPO(
optimizer_cls=SGD, optimizer_kwargs={
“lr”: 0.1, “momentum”: 0.9, “weight_decay”: 1e-4,
},
),
)#
It can be created from the string class import path such as
```python from otx.algo.classification.mobilenet_v3_large import MobileNetV3ForMulticlassCls
- model = MobileNetV3ForMulticlassCls(
num_classes=3, optimizer=OptimizerCallableSupportHPO(
optimizer_cls=”torch.optim.SGD”, optimizer_kwargs={
“lr”: 0.1, “momentum”: 0.9, “weight_decay”: 1e-4,
},
),
)#
- __call__(params: params_t) Optimizer[source]#
Create torch.optim.Optimizer instance for the given parameters.
- classmethod from_callable(func: OptimizerCallable) OptimizerCallableSupportHPO[source]#
Create this class instance from an existing optimizer callable.
- to_lazy_instance() ClassType[source]#
Return lazy instance of this class.
Because OTX is rely on jsonargparse library, the default value of class initialization argument should be the lazy instance. Please refer to https://jsonargparse.readthedocs.io/en/stable/#default-values for more details.
Examples
This is an example to implement a new model with a SGD optimizer and custom configurations as a default.
```python class MyAwesomeMulticlassClsModel(OTXMulticlassClsModel):
- def __init__(
self, label_info: LabelInfoTypes, optimizer: OptimizerCallable = OptimizerCallableSupportHPO(
optimizer_cls=SGD, optimizer_kwargs={
“lr”: 0.1, “momentum”: 0.9, “weight_decay”: 1e-4,
},
).to_lazy_instance(), scheduler: LRSchedulerCallable | LRSchedulerListCallable = DefaultSchedulerCallable, metric: MetricCallable = MultiClassClsMetricCallable, torch_compile: bool = False,
) -> None: …