from typing import Literal
from torch import nn
from torch.optim import Optimizer
from torch_uncertainty.models import mc_dropout
from torch_uncertainty.models.vgg import (
packed_vgg,
vgg,
)
from torch_uncertainty.routines.classification import ClassificationRoutine
from torch_uncertainty.transforms import RepeatTarget
ENSEMBLE_METHODS = ["mc-dropout", "packed"]
[docs]class VGGBaseline(ClassificationRoutine):
versions = {
"std": vgg,
"mc-dropout": vgg,
"packed": packed_vgg,
}
archs = [11, 13, 16, 19]
def __init__(
self,
num_classes: int,
in_channels: int,
loss: nn.Module,
version: Literal["std", "mc-dropout", "packed"],
arch: int,
style: str = "imagenet",
num_estimators: int = 1,
dropout_rate: float = 0.0,
last_layer_dropout: bool = False,
optim_recipe: dict | Optimizer | None = None,
mixup_params: dict | None = None,
groups: int = 1,
alpha: int | None = None,
gamma: int = 1,
ood_criterion: Literal["msp", "logit", "energy", "entropy", "mi", "vr"] = "msp",
log_plots: bool = False,
save_in_csv: bool = False,
calibration_set: Literal["val", "test"] = "val",
eval_ood: bool = False,
eval_shift: bool = False,
eval_grouping_loss: bool = False,
) -> None:
r"""VGG backbone baseline for classification providing support for
various versions and architectures.
Args:
num_classes (int): Number of classes to predict.
in_channels (int): Number of input channels.
loss (nn.Module): Training loss.
optim_recipe (Any): optimization recipe, corresponds to
what expect the `LightningModule.configure_optimizers()
<https://pytorch-lightning.readthedocs.io/en/stable/common/lightning_module.html#configure-optimizers>`_
method.
version (str):
Determines which VGG version to use:
- ``"std"``: original VGG
- ``"mc-dropout"``: Monte Carlo Dropout VGG
- ``"packed"``: Packed-Ensembles VGG
arch (int):
Determines which VGG architecture to use:
- ``11``: VGG-11
- ``13``: VGG-13
- ``16``: VGG-16
- ``19``: VGG-19
style (str, optional): Which VGG style to use. Defaults to
``imagenet``.
num_estimators (int, optional): Number of estimators in the ensemble.
Only used if :attr:`version` is either ``"packed"``, ``"batched"``
or ``"masked"`` Defaults to ``None``.
dropout_rate (float, optional): Dropout rate. Defaults to ``0.0``.
mixup_params (dict, optional): Mixup parameters. Can include mixtype,
mixmode, dist_sim, kernel_tau_max, kernel_tau_std,
mixup_alpha, and cutmix_alpha. If None, no augmentations.
Defaults to ``None``.
last_layer_dropout (bool): whether to apply dropout to the last layer only.
groups (int, optional): Number of groups in convolutions. Defaults to
``1``.
alpha (float, optional): Expansion factor affecting the width of the
estimators. Only used if :attr:`version` is ``"packed"``. Defaults
to ``None``.
gamma (int, optional): Number of groups within each estimator. Only
used if :attr:`version` is ``"packed"`` and scales with
:attr:`groups`. Defaults to ``1s``.
ood_criterion (str, optional): OOD criterion. Defaults to ``"msp"``.
MSP is the maximum softmax probability, logit is the maximum
logit, entropy is the entropy of the mean prediction, mi is the
mutual information of the ensemble and vr is the variation ratio
of the ensemble.
log_plots (bool, optional): Indicates whether to log the plots or not.
Defaults to ``False``.
save_in_csv (bool, optional): Indicates whether to save the results in
a csv file or not. Defaults to ``False``.
calibration_set (Callable, optional): Calibration set. Defaults to
``None``.
eval_ood (bool, optional): Indicates whether to evaluate the
OOD detection or not. Defaults to ``False``.
eval_shift (bool): Whether to evaluate on shifted data. Defaults to
``False``.
eval_grouping_loss (bool, optional): Indicates whether to evaluate the
grouping loss or not. Defaults to ``False``.
Raises:
ValueError: If :attr:`version` is not either ``"std"``,
``"packed"``, ``"batched"`` or ``"masked"``.
Returns:
LightningModule: VGG baseline ready for training and evaluation.
"""
params = {
"dropout_rate": dropout_rate,
"in_channels": in_channels,
"num_classes": num_classes,
"style": style,
"groups": groups,
"arch": arch,
}
if version not in self.versions:
raise ValueError(f"Unknown version: {version}")
format_batch_fn = nn.Identity()
if version == "std":
params |= {
"dropout_rate": dropout_rate,
}
elif version == "mc-dropout":
params |= {
"dropout_rate": dropout_rate,
"num_estimators": num_estimators,
}
if version in ENSEMBLE_METHODS:
params |= {
"num_estimators": num_estimators,
}
if version != "mc-dropout":
format_batch_fn = RepeatTarget(num_repeats=num_estimators)
if version == "packed":
params |= {
"alpha": alpha,
"style": style,
"gamma": gamma,
}
if version == "mc-dropout": # std VGGs don't have `num_estimators`
del params["num_estimators"]
model = self.versions[version](**params)
if version == "mc-dropout":
model = mc_dropout(
model=model,
num_estimators=num_estimators,
last_layer=last_layer_dropout,
)
super().__init__(
num_classes=num_classes,
model=model,
loss=loss,
is_ensemble=version in ENSEMBLE_METHODS,
format_batch_fn=format_batch_fn,
optim_recipe=optim_recipe,
mixup_params=mixup_params,
eval_ood=eval_ood,
eval_shift=eval_shift,
ood_criterion=ood_criterion,
log_plots=log_plots,
save_in_csv=save_in_csv,
calibration_set=calibration_set,
eval_grouping_loss=eval_grouping_loss,
)
self.save_hyperparameters(ignore=["loss"])
