Source code for annotlib.classifier_based

import numpy as np
import copy

from annotlib.standard import StandardAnnot
from annotlib.utils import check_labelling_array, check_shape

from sklearn.base import is_classifier
from sklearn.utils import check_X_y, check_array
from sklearn.preprocessing import LabelEncoder
from sklearn.svm import SVC


[docs]class ClassifierBasedAnnot(StandardAnnot): """ClassifierBasedAnnot Annotators can be seen as human classifiers. Hence, we use classifiers based on machine learning techniques to represent these annotators. Given a data set comprising samples with their true labels, a classifier is trained on a subset of sample-label-pairs. Subsequently, this trained classifier is used as proxy of an annotator. As a result, the labels for a sample are provided by this classifier as well as the confidence scores which are the posterior probability estimates for the predicted class label. Parameters ---------- X: array-like, shape (n_samples, n_features) Samples of the whole data set. y_true: array-like, shape (n_samples) True class labels of the given samples X. n_annotators: int Number of annotators who are simulated. classifiers: sklearn.base.ClassifierMixin | list of ClassifierMixin, shape (n_annotators) The classifiers parameter is either a single sklearn classifier supporting :py:method::predict_proba` or a list of such classifiers. If the parameter is not a list, each annotator is simulated on the same classification model, whereas if it is a list, the annotators may be simulated on different classifier types or even different parametrisations. The default classifiers parameter is a list of SVMs with the same parameters. train_ratios: 'one-hot' | 'equidistant' | array-like, shape (n_annotators, n_classes) The entry `train_ratios_[j, i]` indicates the ratio of samples of class i used for training the classifier of annotator j, e.g. `train_ratios_[2,4]=0.3`: 30% of the samples for class 4 are used to train the classifier of annotator with the id 2. features: array-like, boolean, shape (n_annotators, n_features) This parameter is a boolean array indicating which features are considered for the training of an annotator's classifier, e.g. features_[a] = [0, 0, 1] means the classifier of annotator a is trained with the last of the three available features. confidence_noise: array-like, shape (n_annotators) An entry of confidence_noise defines the interval from which the noise is uniformly drawn, e.g. confidence_noise[a] = 0.2 results in sampling n_samples times from U(-0.2, 0.2) and adding this noise to the confidence scores. Zero noise is the default value for each annotator. random_state: None | int | numpy.random.RandomState The random state used for generating class labels of the annotators. Attributes ---------- X_: numpy.ndarray, shape (n_samples, n_features) Samples of the whole data set. Y_: numpy.ndarray, shape (n_samples, n_annotators) Class labels of the given samples X. C_: numpy.ndarray, shape (n_samples, n_annotators) confidence score for labelling the given samples x. C_noise_: numpy.ndarray, shape (n_samples, n_annotators) The uniformly noise for each annotator and each sample, e.g. C[x_idx, a_idx] indicates the noise for the confidence score of annotator with id a_idx in labelling sample with id x_idx. n_annotators_: int Number of annotators. n_queries_: numpy.ndarray, shape (n_annotators) An entry n_queries_[a_idx] indicates how many queries annotator with id a_idx has processed. queried_flags_: numpy.ndarray, shape (n_samples, n_annotators) An entry queried_flags_[a_idx, x_idx] is a boolean indicating whether annotator with id a_idx has provided a class label for sample with id x_idx. y_true_: numpy.ndarray, shape (n_samples) The true class labels of the given samples. train_ratios_: numpy.ndarray, shape (n_annotators, n_classes) The entry `train_ratios_[j, i]` indicates the ratio of samples of class i used for training the classifier of annotator j, e.g. `train_ratios_[2,4]=0.3`: 30% of the samples for class 4 are used to train the classifier of annotator 2. classifiers_: list of sklearn.base.ClassifierMixin, shape (n_annotators) The fitted classification models of the annotators. features_: array-numpy.ndarray, boolean, shape (n_annotators, n_features) This parameter is a boolean array indicating which features are considered for the training of an annotator's classifier, e.g. features_[a] = [0, 0, 1] means the classifier of annotator a is trained with the last of the three available features. random_state_: None | int | numpy.random.RandomState The random state used for generating class labels of the annotators. Examples -------- >>> from sklearn.datasets import load_iris >>> from sklearn.gaussian_process import GaussianProcessClassifier >>> from sklearn.svm import SVC >>> # load iris data set >>> X, y_true = load_iris(return_X_y=True) >>> # create list of SVM and Gaussian Process classifier >>> classifiers = [SVC(C=1, probability=True, gamma='auto'), SVC(C=3, probability=True), GaussianProcessClassifier()] >>> # simulate annotators on the iris data set >>> annotators = ClassifierBasedAnnot(X=X, y_true=y_true, classifiers=classifiers, n_annotators=3) >>> # the number of annotators must be equal to the number of classifiers >>> annotators.n_annotators() 3 >>> # query class labels of 100 samples from annotators a_0, a_2 >>> annotators.class_labels(X=X[0:100], y_true=y_true[0:100], annotator_ids=[0, 2], query_value=100).shape (100, 3) >>> # check query values >>> annotators.n_queries() array([100, 0, 100]) >>> # query confidence scores of these 100 samples from annotators a_0, a_2 >>> annotators.confidence_scores(X=X[0:100], y_true=y_true[0:100], annotator_ids=[0, 2]).shape (100, 3) >>> # query values are not affected by calling the confidence score method >>> annotators.n_queries() array([100, 0, 100]) """ def __init__(self, X, y_true, classifiers=None, n_annotators=5, train_ratios='one-hot', features=None, confidence_noise=None, random_state=None): # check shape of samples and labels self.X_, self.y_true_ = check_X_y(X, y_true) n_samples = np.size(self.X_, 0) n_features = np.size(self.X_, 1) # check and set number of annotators, query number and queried samples self._check_parameters(n_annotators, n_samples, confidence_noise, random_state) # initialise container for class labels and confidence scores self.Y_ = np.empty((n_samples, self.n_annotators_)) self.C_ = np.empty((n_samples, self.n_annotators_)) # transform class labels to interval [0, n_classes-1] le = LabelEncoder().fit(self.y_true_) y_transformed = le.transform(self.y_true_) y_unique = np.unique(y_transformed) n_classes = len(y_unique) # check classifier models if not isinstance(classifiers, list): clf = SVC(random_state=self.random_state_, probability=True, gamma='auto') if classifiers is None else classifiers classifiers = [copy.deepcopy(clf) for _ in range(self.n_annotators())] for clf in classifiers: if len(classifiers) != self.n_annotators() or not is_classifier(clf) or getattr(clf, 'predict_proba', None) is None: raise TypeError('The parameter `classifiers` must be a single sklearn classifier or a list of sklearn ' 'classifiers supporting the method :py:method::`predict_proba`.') self.classifiers_ = classifiers # check shape and values of label_acc parameter self.train_ratios_ = train_ratios if isinstance(train_ratios, str): if train_ratios == 'one-hot': train_ratios = np.empty((self.n_annotators(), n_classes)) class_indices = np.arange(0, n_classes) for a_idx in range(self.n_annotators()): class_j = a_idx % n_classes train_ratios[a_idx, class_j] = 1 train_ratios[a_idx, class_indices != class_j] = 0.2 elif train_ratios == 'equidistant': train_ratios = np.fromfunction(lambda j, i: (j + 1) * (1 / (n_annotators)), (n_annotators, n_classes), dtype=int) self.train_ratios_ = check_labelling_array(train_ratios, (self.n_annotators(), n_classes), 'train_ratios') # check features flag array self.features_ = np.full((self.n_annotators(), n_features), True) if features is None else features self.features_ = check_array(self.features_) self.features_ = check_shape(self.features_, (self.n_annotators(), n_features), parameter_name='features') if self.features_.dtype != np.dtype('bool'): raise TypeError('The parameter `features` must be a boolean array.') # container for generated class labels and confidence scores of simulated annotators class_indices = [np.where(y_transformed == c)[0] for c in y_unique] # generate class labels depending on the clustering and the corresponding labelling accuracies for a_idx in range(n_annotators): train_size = [int(self.train_ratios_[a_idx, c] * len(class_indices[c]) + .5) for c in y_unique] train = [self.random_state_.choice(class_indices[c], size=train_size[c], replace=False) for c in y_unique] train = np.hstack(train) X_train = self.X_[train] X_train = X_train[:, self.features_[a_idx]] y_train = y_transformed[train] self.classifiers_[a_idx] = self.classifiers_[a_idx].fit(X_train, y_train) y_predict = self.classifiers_[a_idx].predict(self.X_[:, self.features_[a_idx]]) self.Y_[:, a_idx] = le.inverse_transform(y_predict) self.C_[:, a_idx] = np.max(self.classifiers_[a_idx].predict_proba(self.X_[:, self.features_[a_idx]]), axis=1) self._add_confidence_noise(probabilistic=True)