"""
.. image:: https://img.shields.io/badge/classification-yes-brightgreen?style=flat-square
:alt: classification badge
.. image:: https://img.shields.io/badge/segmentation-yes-brightgreen?style=flat-square
:alt: classification badge
.. autoclass:: pytorch_ood.detector.MaxSoftmax
:members:
:exclude-members: fit, fit_features
"""
import logging
from typing import Optional, TypeVar
import torch.nn
from torch import Tensor, tensor
from torch.nn import Module
from torch.nn.functional import nll_loss
from torch.optim import LBFGS
from torch.utils.data import DataLoader
from pytorch_ood.utils import extract_features, is_known
from ..api import Detector, ModelNotSetException, RequiresFittingException
log = logging.getLogger(__name__)
Self = TypeVar("Self")
[docs]
class MaxSoftmax(Detector):
"""
Implements the Maximum Softmax Probability (MSP) Thresholding baseline for OOD detection.
Optionally, implements temperature scaling, which divides the logits by a constant temperature :math:`T`
before calculating the softmax. The score is calculated as:
.. math:: - \\max_y \\sigma_y(f(x) / T)
where :math:`\\sigma` is the softmax function and :math:`\\sigma_y` indicates the :math:`y^{th}` value of the
resulting probability vector.
:see Paper:
`ArXiv <https://arxiv.org/abs/1610.02136>`_
:see Implementation:
`GitHub <https://github.com/hendrycks/error-detection>`_
"""
def __init__(self, model: Module, t: Optional[float] = 1.0):
"""
:param model: neural network to use
:param t: temperature value :math:`T`. Default is 1.
"""
super(MaxSoftmax, self).__init__()
self.t = tensor(t)
self.model = model
[docs]
def predict(self, x: Tensor) -> Tensor:
"""
:param x: input, will be passed through model
"""
if self.model is None:
raise ModelNotSetException
return self.predict_features(self.model(x))
def fit(self: Self, *args, **kwargs) -> Self:
"""
Not required
"""
return self
def fit_features(self: Self, *args, **kwargs) -> Self:
"""
Not required
"""
return self
[docs]
def predict_features(self, logits: Tensor) -> Tensor:
"""
:param logits: logits given by the model
"""
return MaxSoftmax.score(logits, self.t)
[docs]
@staticmethod
def score(logits: Tensor, t: Optional[float] = 1.0) -> Tensor:
"""
:param logits: logits for samples
:param t: temperature value
"""
return -logits.div(t).softmax(dim=1).max(dim=1).values