Source code for pyrelational.strategies.classification.classification_strategy

"""ClassificationStrategy class for active learning in classification tasks."""

import math
from typing import Any, List

import torch
from torch import Tensor

from pyrelational.data_managers import DataManager
from pyrelational.model_managers import ModelManager
from pyrelational.strategies.abstract_strategy import Strategy


[docs] class ClassificationStrategy(Strategy): """A base active learning strategy class for classification."""
[docs] def __call__( self, num_annotate: int, data_manager: DataManager, model_manager: ModelManager[Any, Any] ) -> List[int]: """ Identify samples for labelling based on user defined scoring and sampling function. :param num_annotate: number of samples to annotate :param data_manager: A pyrelational data manager which keeps track of what has been labelled and creates data loaders for active learning :param model_manager: A pyrelational model manager which wraps a user defined ML model to handle instantiation, training, testing, as well as uncertainty quantification :return: list of indices to annotate """ output = self.train_and_infer(data_manager=data_manager, model_manager=model_manager).mean(0) if not torch.allclose(output.sum(1), torch.tensor(1.0)): output = softmax(output) uncertainty = self.scorer(output) return self.sampler(uncertainty, data_manager.u_indices, num_annotate)
[docs] def softmax(scores: Tensor, base: float = math.e, axis: int = -1) -> Tensor: """Return softmax array for array of scores. Converts a set of raw scores from a model (logits) into a probability distribution via softmax. The probability distribution will be a set of real numbers such that each is in the range 0-1.0 and the sum is 1.0. Assumes input is a pytorch tensor: tensor([1.0, 4.0, 2.0, 3.0]) :param scores: (pytorch tensor) a pytorch tensor of any positive/negative real numbers. :param base: the base for the exponential (default e) :param: axis to apply softmax on scores :return: tensor of softmaxed scores """ exps = base ** scores.float() # exponential for each value in array sum_exps = torch.sum(exps, dim=axis, keepdim=True) # sum of all exponentials prob_dist: Tensor = exps / sum_exps # normalize exponentials return prob_dist