cardinal.uncertainty¶
Functions¶
- cardinal.uncertainty.confidence_score(classifier, X: ndarray) ndarray[source]¶
Measure the confidence score of a model for a set of samples.
- Parameters
classifier – The classifier for which the labels are to be queried.
X – The pool of samples to query from.
- Returns
The confidence score for each sample.
- cardinal.uncertainty.entropy_score(classifier, X: ndarray) ndarray[source]¶
Entropy sampling query strategy, uses entropy of all probabilities as score.
This strategy selects the samples with the highest entropy in their prediction probabilities.
- Parameters
classifier – The classifier for which the labels are to be queried.
X – The pool of samples to query from.
n_instances – Number of samples to be queried.
- Returns
The entropy score for each label
- cardinal.uncertainty.margin_score(classifier, X: ndarray) ndarray[source]¶
Compute the difference between the two top probability classes for each sample.
This strategy takes the probabilities of top two classes and uses their difference as a score for selection.
- Parameters
classifier – The classifier for which the labels are to be queried.
X – The pool of samples to query from.
- Returns
The margin score for each sample.
Classes¶
- class cardinal.uncertainty.ConfidenceSampler(classifier, batch_size: int, strategy: str = 'top', assume_fitted: bool = False, verbose: int = 0)[source]¶
Selects samples with lowest prediction confidence.
Lowest confidence sampling looks at the probability of the class predicted by the classifier and selects the samples where this probability is the lowest.
- Parameters
classifier – Classifier used to determine the prediction confidence. The object must comply with scikit-learn interface and expose a predict_proba method.
batch_size – Number of samples to draw when predicting.
assume_fitted – If true, classifier is not refit
verbose – The verbosity level. Defaults to 0.
- classifier_¶
The fitted classifier.
- fit(X: array, y: array) ConfidenceSampler[source]¶
Fit the estimator on labeled samples.
- Parameters
X – Labeled samples of shape (n_samples, n_features).
y – Labels of shape (n_samples).
- Returns
The object itself
- class cardinal.uncertainty.EntropySampler(classifier, batch_size: int, strategy: str = 'top', assume_fitted: bool = False, verbose: int = 0)[source]¶
Selects samples with greatest entropy among all class probabilities.
Greatest entropy sampling measures the uncertainty of the model over all classes through the entropy of the probabilites of all classes. Highest entropy samples are selected.
- Parameters
classifier – Classifier used to determine the prediction confidence. The object must comply with scikit-learn interface and expose a predict_proba method.
batch_size – Number of samples to draw when predicting.
assume_fitted – If true, classifier is not refit
verbose – The verbosity level. Defaults to 0.
- classifier_¶
The fitted classifier.
- fit(X: array, y: array) EntropySampler[source]¶
Fit the estimator on labeled samples.
- Parameters
X – Labeled samples of shape (n_samples, n_features).
y – Labels of shape (n_samples).
- Returns
The object itself
- class cardinal.uncertainty.MarginSampler(classifier, batch_size: int, strategy: str = 'top', assume_fitted: bool = False, verbose: int = 0)[source]¶
Selects samples with greatest confusion between the top two classes.
Smallest margin sampling uses the difference of predicted probability between the top two classes to select the samples on which the model is hesitating the most, hence the lowest difference.
- Parameters
classifier – Classifier used to determine the prediction confidence. The object must comply with scikit-learn interface and expose a predict_proba method.
batch_size – Number of samples to draw when predicting.
assume_fitted – If true, classifier is not refit
verbose – The verbosity level. Defaults to 0.
- classifier_¶
The fitted classifier.
- fit(X: array, y: array) MarginSampler[source]¶
Fit the estimator on labeled samples.
- Parameters
X – Labeled samples of shape (n_samples, n_features).
y – Labels of shape (n_samples).
- Returns
The object itself