adding documentation

This commit is contained in:
Alejandro Moreo Fernandez 2021-12-15 15:27:43 +01:00
parent 5deb92b457
commit 3835f89e9d
10 changed files with 1944 additions and 406 deletions

View File

@ -24,6 +24,10 @@ Do we want to cover cross-lingual quantification natively in QuaPy, or does it m
Current issues: Current issues:
========================================== ==========================================
Revise the class structure of quantification methods and the methods they inherit... There is some confusion regarding
methods isbinary, isprobabilistic, and the like. The attribute "learner_" in aggregative quantifiers is also
confusing, since there is a getter and a setter.
Remove the "deep" in get_params. There is no real compatibility with scikit-learn as for now.
SVMperf-based learners do not remove temp files in __del__? SVMperf-based learners do not remove temp files in __del__?
In binary quantification (hp, kindle, imdb) we used F1 in the minority class (which in kindle and hp happens to be the In binary quantification (hp, kindle, imdb) we used F1 in the minority class (which in kindle and hp happens to be the
negative class). This is not covered in this new implementation, in which the binary case is not treated as such, but as negative class). This is not covered in this new implementation, in which the binary case is not treated as such, but as

View File

@ -80,8 +80,6 @@
<li><a href="quapy.html#quapy.error.acc_error">acc_error() (in module quapy.error)</a> <li><a href="quapy.html#quapy.error.acc_error">acc_error() (in module quapy.error)</a>
</li> </li>
<li><a href="quapy.html#quapy.error.acce">acce() (in module quapy.error)</a> <li><a href="quapy.html#quapy.error.acce">acce() (in module quapy.error)</a>
</li>
<li><a href="quapy.method.html#quapy.method.meta.Ensemble.accuracy_policy">accuracy_policy() (quapy.method.meta.Ensemble method)</a>
</li> </li>
<li><a href="quapy.data.html#quapy.data.preprocessing.IndexTransformer.add_word">add_word() (quapy.data.preprocessing.IndexTransformer method)</a> <li><a href="quapy.data.html#quapy.data.preprocessing.IndexTransformer.add_word">add_word() (quapy.data.preprocessing.IndexTransformer method)</a>
</li> </li>
@ -226,12 +224,6 @@
<li><a href="quapy.method.html#quapy.method.neural.QuaNetTrainer.clean_checkpoint_dir">clean_checkpoint_dir() (quapy.method.neural.QuaNetTrainer method)</a> <li><a href="quapy.method.html#quapy.method.neural.QuaNetTrainer.clean_checkpoint_dir">clean_checkpoint_dir() (quapy.method.neural.QuaNetTrainer method)</a>
</li> </li>
<li><a href="quapy.classification.html#quapy.classification.neural.CNNnet">CNNnet (class in quapy.classification.neural)</a> <li><a href="quapy.classification.html#quapy.classification.neural.CNNnet">CNNnet (class in quapy.classification.neural)</a>
</li>
<li><a href="quapy.method.html#quapy.method.aggregative.ThresholdOptimization.compute_fpr">compute_fpr() (quapy.method.aggregative.ThresholdOptimization method)</a>
</li>
<li><a href="quapy.method.html#quapy.method.aggregative.ThresholdOptimization.compute_table">compute_table() (quapy.method.aggregative.ThresholdOptimization method)</a>
</li>
<li><a href="quapy.method.html#quapy.method.aggregative.ThresholdOptimization.compute_tpr">compute_tpr() (quapy.method.aggregative.ThresholdOptimization method)</a>
</li> </li>
<li><a href="quapy.data.html#quapy.data.base.LabelledCollection.counts">counts() (quapy.data.base.LabelledCollection method)</a> <li><a href="quapy.data.html#quapy.data.base.LabelledCollection.counts">counts() (quapy.data.base.LabelledCollection method)</a>
</li> </li>
@ -257,6 +249,8 @@
</ul></li> </ul></li>
<li><a href="quapy.classification.html#quapy.classification.neural.TextClassifierNet.dimensions">dimensions() (quapy.classification.neural.TextClassifierNet method)</a> <li><a href="quapy.classification.html#quapy.classification.neural.TextClassifierNet.dimensions">dimensions() (quapy.classification.neural.TextClassifierNet method)</a>
</li> </li>
</ul></td>
<td style="width: 33%; vertical-align: top;"><ul>
<li><a href="quapy.classification.html#quapy.classification.neural.CNNnet.document_embedding">document_embedding() (quapy.classification.neural.CNNnet method)</a> <li><a href="quapy.classification.html#quapy.classification.neural.CNNnet.document_embedding">document_embedding() (quapy.classification.neural.CNNnet method)</a>
<ul> <ul>
@ -265,15 +259,9 @@
<li><a href="quapy.classification.html#quapy.classification.neural.TextClassifierNet.document_embedding">(quapy.classification.neural.TextClassifierNet method)</a> <li><a href="quapy.classification.html#quapy.classification.neural.TextClassifierNet.document_embedding">(quapy.classification.neural.TextClassifierNet method)</a>
</li> </li>
</ul></li> </ul></li>
</ul></td>
<td style="width: 33%; vertical-align: top;"><ul>
<li><a href="quapy.html#quapy.util.download_file">download_file() (in module quapy.util)</a> <li><a href="quapy.html#quapy.util.download_file">download_file() (in module quapy.util)</a>
</li> </li>
<li><a href="quapy.html#quapy.util.download_file_if_not_exists">download_file_if_not_exists() (in module quapy.util)</a> <li><a href="quapy.html#quapy.util.download_file_if_not_exists">download_file_if_not_exists() (in module quapy.util)</a>
</li>
<li><a href="quapy.method.html#quapy.method.meta.Ensemble.ds_policy">ds_policy() (quapy.method.meta.Ensemble method)</a>
</li>
<li><a href="quapy.method.html#quapy.method.meta.Ensemble.ds_policy_get_posteriors">ds_policy_get_posteriors() (quapy.method.meta.Ensemble method)</a>
</li> </li>
</ul></td> </ul></td>
</tr></table> </tr></table>
@ -619,7 +607,7 @@
<ul> <ul>
<li><a href="quapy.data.html#quapy.data.base.LabelledCollection.n_classes">(quapy.data.base.LabelledCollection property)</a> <li><a href="quapy.data.html#quapy.data.base.LabelledCollection.n_classes">(quapy.data.base.LabelledCollection property)</a>
</li> </li>
<li><a href="quapy.method.html#quapy.method.aggregative.AggregativeQuantifier.n_classes">(quapy.method.aggregative.AggregativeQuantifier property)</a> <li><a href="quapy.method.html#quapy.method.base.BaseQuantifier.n_classes">(quapy.method.base.BaseQuantifier property)</a>
</li> </li>
</ul></li> </ul></li>
<li><a href="quapy.html#quapy.evaluation.natural_prevalence_prediction">natural_prevalence_prediction() (in module quapy.evaluation)</a> <li><a href="quapy.html#quapy.evaluation.natural_prevalence_prediction">natural_prevalence_prediction() (in module quapy.evaluation)</a>
@ -650,14 +638,6 @@
<td style="width: 33%; vertical-align: top;"><ul> <td style="width: 33%; vertical-align: top;"><ul>
<li><a href="quapy.method.html#quapy.method.aggregative.OneVsAll">OneVsAll (class in quapy.method.aggregative)</a> <li><a href="quapy.method.html#quapy.method.aggregative.OneVsAll">OneVsAll (class in quapy.method.aggregative)</a>
</li> </li>
<li><a href="quapy.method.html#quapy.method.aggregative.MS.optimize_threshold">optimize_threshold() (quapy.method.aggregative.MS method)</a>
<ul>
<li><a href="quapy.method.html#quapy.method.aggregative.MS2.optimize_threshold">(quapy.method.aggregative.MS2 method)</a>
</li>
<li><a href="quapy.method.html#quapy.method.aggregative.ThresholdOptimization.optimize_threshold">(quapy.method.aggregative.ThresholdOptimization method)</a>
</li>
</ul></li>
</ul></td> </ul></td>
</tr></table> </tr></table>
@ -721,8 +701,6 @@
<li><a href="quapy.method.html#quapy.method.aggregative.ProbabilisticAdjustedClassifyAndCount">ProbabilisticAdjustedClassifyAndCount (in module quapy.method.aggregative)</a> <li><a href="quapy.method.html#quapy.method.aggregative.ProbabilisticAdjustedClassifyAndCount">ProbabilisticAdjustedClassifyAndCount (in module quapy.method.aggregative)</a>
</li> </li>
<li><a href="quapy.method.html#quapy.method.aggregative.ProbabilisticClassifyAndCount">ProbabilisticClassifyAndCount (in module quapy.method.aggregative)</a> <li><a href="quapy.method.html#quapy.method.aggregative.ProbabilisticClassifyAndCount">ProbabilisticClassifyAndCount (in module quapy.method.aggregative)</a>
</li>
<li><a href="quapy.method.html#quapy.method.meta.Ensemble.ptr_policy">ptr_policy() (quapy.method.meta.Ensemble method)</a>
</li> </li>
</ul></td> </ul></td>
</tr></table> </tr></table>
@ -968,11 +946,11 @@
</ul></li> </ul></li>
</ul></td> </ul></td>
<td style="width: 33%; vertical-align: top;"><ul> <td style="width: 33%; vertical-align: top;"><ul>
<li><a href="quapy.method.html#quapy.method.aggregative.SLD">SLD (in module quapy.method.aggregative)</a>
</li>
<li><a href="quapy.html#quapy.error.smooth">smooth() (in module quapy.error)</a> <li><a href="quapy.html#quapy.error.smooth">smooth() (in module quapy.error)</a>
</li> </li>
<li><a href="quapy.method.html#quapy.method.aggregative.ACC.solve_adjustment">solve_adjustment() (quapy.method.aggregative.ACC class method)</a> <li><a href="quapy.method.html#quapy.method.aggregative.ACC.solve_adjustment">solve_adjustment() (quapy.method.aggregative.ACC class method)</a>
</li>
<li><a href="quapy.method.html#quapy.method.meta.Ensemble.sout">sout() (quapy.method.meta.Ensemble method)</a>
</li> </li>
<li><a href="quapy.data.html#quapy.data.base.LabelledCollection.split_stratified">split_stratified() (quapy.data.base.LabelledCollection method)</a> <li><a href="quapy.data.html#quapy.data.base.LabelledCollection.split_stratified">split_stratified() (quapy.data.base.LabelledCollection method)</a>
</li> </li>
@ -1013,14 +991,12 @@
<li><a href="quapy.data.html#quapy.data.preprocessing.text2tfidf">text2tfidf() (in module quapy.data.preprocessing)</a> <li><a href="quapy.data.html#quapy.data.preprocessing.text2tfidf">text2tfidf() (in module quapy.data.preprocessing)</a>
</li> </li>
<li><a href="quapy.classification.html#quapy.classification.neural.TextClassifierNet">TextClassifierNet (class in quapy.classification.neural)</a> <li><a href="quapy.classification.html#quapy.classification.neural.TextClassifierNet">TextClassifierNet (class in quapy.classification.neural)</a>
</li>
<li><a href="quapy.method.html#quapy.method.aggregative.ThresholdOptimization">ThresholdOptimization (class in quapy.method.aggregative)</a>
</li> </li>
</ul></td> </ul></td>
<td style="width: 33%; vertical-align: top;"><ul> <td style="width: 33%; vertical-align: top;"><ul>
<li><a href="quapy.classification.html#quapy.classification.neural.TorchDataset">TorchDataset (class in quapy.classification.neural)</a> <li><a href="quapy.method.html#quapy.method.aggregative.ThresholdOptimization">ThresholdOptimization (class in quapy.method.aggregative)</a>
</li> </li>
<li><a href="quapy.method.html#quapy.method.aggregative.training_helper">training_helper() (in module quapy.method.aggregative)</a> <li><a href="quapy.classification.html#quapy.classification.neural.TorchDataset">TorchDataset (class in quapy.classification.neural)</a>
</li> </li>
<li><a href="quapy.classification.html#quapy.classification.methods.LowRankLogisticRegression.transform">transform() (quapy.classification.methods.LowRankLogisticRegression method)</a> <li><a href="quapy.classification.html#quapy.classification.methods.LowRankLogisticRegression.transform">transform() (quapy.classification.methods.LowRankLogisticRegression method)</a>

Binary file not shown.

File diff suppressed because it is too large Load Diff

File diff suppressed because one or more lines are too long

View File

@ -23,46 +23,109 @@ from quapy.method.base import BaseQuantifier, BinaryQuantifier
class AggregativeQuantifier(BaseQuantifier): class AggregativeQuantifier(BaseQuantifier):
""" """
Abstract class for quantification methods that base their estimations on the aggregation of classification Abstract class for quantification methods that base their estimations on the aggregation of classification
results. Aggregative Quantifiers thus implement a _classify_ method and maintain a _learner_ attribute. results. Aggregative Quantifiers thus implement a :meth:`classify` method and maintain a :attr:`learner` attribute.
Subclasses of this abstract class must implement the method :meth:`aggregate` which computes the aggregation
of label predictions. The method :meth:`quantify` comes with a default implementation based on
:meth:`classify` and :meth:`aggregate`.
""" """
@abstractmethod @abstractmethod
def fit(self, data: LabelledCollection, fit_learner=True): ... def fit(self, data: LabelledCollection, fit_learner=True):
"""
Trains the aggregative quantifier
:param data: a :class:`quapy.data.base.LabelledCollection` consisting of the training data
:param fit_learner: whether or not to train the learner (default is True). Set to False if the
learner has been trained outside the quantifier.
:return: self
"""
...
@property @property
def learner(self): def learner(self):
"""
Gives access to the classifier
:return: the classifier (typically an sklearn's Estimator)
"""
return self.learner_ return self.learner_
@learner.setter @learner.setter
def learner(self, value): def learner(self, classifier):
self.learner_ = value """
Setter for the classifier
:param classifier: the classifier
"""
self.learner_ = classifier
def classify(self, instances): def classify(self, instances):
"""
Provides the label predictions for the given instances.
:param instances: array-like
:return: np.ndarray of shape `(n_instances,)` with label predictions
"""
return self.learner.predict(instances) return self.learner.predict(instances)
def quantify(self, instances): def quantify(self, instances):
"""
Generate class prevalence estimates for the sample's instances by aggregating the label predictions generated
by the classifier.
:param instances: array-like
:return: `np.ndarray` of shape `(self.n_classes_,)` with class prevalence estimates.
"""
classif_predictions = self.classify(instances) classif_predictions = self.classify(instances)
return self.aggregate(classif_predictions) return self.aggregate(classif_predictions)
@abstractmethod @abstractmethod
def aggregate(self, classif_predictions: np.ndarray): ... def aggregate(self, classif_predictions: np.ndarray):
"""
Implements the aggregation of label predictions.
:param classif_predictions: `np.ndarray` of label predictions
:return: `np.ndarray` of shape `(self.n_classes_,)` with class prevalence estimates.
"""
...
def get_params(self, deep=True): def get_params(self, deep=True):
"""
Return the current parameters of the quantifier.
:param deep: for compatibility with sklearn
:return: a dictionary of param-value pairs
"""
return self.learner.get_params() return self.learner.get_params()
def set_params(self, **parameters): def set_params(self, **parameters):
"""
Set the parameters of the quantifier.
:param parameters: dictionary of param-value pairs
"""
self.learner.set_params(**parameters) self.learner.set_params(**parameters)
@property
def n_classes(self):
return len(self.classes_)
@property @property
def classes_(self): def classes_(self):
"""
Class labels, in the same order in which class prevalence values are to be computed.
This default implementation actually returns the class labels of the learner.
:return: array-like
"""
return self.learner.classes_ return self.learner.classes_
@property @property
def aggregative(self): def aggregative(self):
"""
Returns True, indicating the quantifier is of type aggregative.
:return: True
"""
return True return True
@ -96,13 +159,14 @@ class AggregativeProbabilisticQuantifier(AggregativeQuantifier):
# Helper # Helper
# ------------------------------------ # ------------------------------------
def training_helper(learner, def _training_helper(learner,
data: LabelledCollection, data: LabelledCollection,
fit_learner: bool = True, fit_learner: bool = True,
ensure_probabilistic=False, ensure_probabilistic=False,
val_split: Union[LabelledCollection, float] = None): val_split: Union[LabelledCollection, float] = None):
""" """
Training procedure common to all Aggregative Quantifiers. Training procedure common to all Aggregative Quantifiers.
:param learner: the learner to be fit :param learner: the learner to be fit
:param data: the data on which to fit the learner. If requested, the data will be split before fitting the learner. :param data: the data on which to fit the learner. If requested, the data will be split before fitting the learner.
:param fit_learner: whether or not to fit the learner (if False, then bypasses any action) :param fit_learner: whether or not to fit the learner (if False, then bypasses any action)
@ -154,8 +218,10 @@ def training_helper(learner,
# ------------------------------------ # ------------------------------------
class CC(AggregativeQuantifier): class CC(AggregativeQuantifier):
""" """
The most basic Quantification method. One that simply classifies all instances and countes how many have been The most basic Quantification method. One that simply classifies all instances and counts how many have been
attributed each of the classes in order to compute class prevalence estimates. attributed to each of the classes in order to compute class prevalence estimates.
:param learner: a sklearn's Estimator that generates a classifier
""" """
def __init__(self, learner: BaseEstimator): def __init__(self, learner: BaseEstimator):
@ -163,19 +229,40 @@ class CC(AggregativeQuantifier):
def fit(self, data: LabelledCollection, fit_learner=True): def fit(self, data: LabelledCollection, fit_learner=True):
""" """
Trains the Classify & Count method unless _fit_learner_ is False, in which case it is assumed to be already fit. Trains the Classify & Count method unless `fit_learner` is False, in which case, the classifier is assumed to
:param data: training data be already fit and there is nothing else to do.
:param data: a :class:`quapy.data.base.LabelledCollection` consisting of the training data
:param fit_learner: if False, the classifier is assumed to be fit :param fit_learner: if False, the classifier is assumed to be fit
:return: self :return: self
""" """
self.learner, _ = training_helper(self.learner, data, fit_learner) self.learner, _ = _training_helper(self.learner, data, fit_learner)
return self return self
def aggregate(self, classif_predictions): def aggregate(self, classif_predictions: np.ndarray):
"""
Computes class prevalence estimates by counting the prevalence of each of the predicted labels.
:param classif_predictions: array-like with label predictions
:return: `np.ndarray` of shape `(self.n_classes_,)` with class prevalence estimates.
"""
return F.prevalence_from_labels(classif_predictions, self.classes_) return F.prevalence_from_labels(classif_predictions, self.classes_)
class ACC(AggregativeQuantifier): class ACC(AggregativeQuantifier):
"""
`Adjusted Classify & Count <https://link.springer.com/article/10.1007/s10618-008-0097-y>`_,
the "adjusted" variant of :class:`CC`, that corrects the predictions of CC
according to the `misclassification rates`.
:param learner: a sklearn's Estimator that generates a classifier
:param val_split: indicates the proportion of data to be used as a stratified held-out validation set in which the
misclassification rates are to be estimated.
This parameter can be indicated as a real value (between 0 and 1, default 0.4), representing a proportion of
validation data, or as an integer, indicating that the misclassification rates should be estimated via
`k`-fold cross validation (this integer stands for the number of folds `k`), or as a
:class:`quapy.data.base.LabelledCollection` (the split itself).
"""
def __init__(self, learner: BaseEstimator, val_split=0.4): def __init__(self, learner: BaseEstimator, val_split=0.4):
self.learner = learner self.learner = learner
@ -183,13 +270,14 @@ class ACC(AggregativeQuantifier):
def fit(self, data: LabelledCollection, fit_learner=True, val_split: Union[float, int, LabelledCollection] = None): def fit(self, data: LabelledCollection, fit_learner=True, val_split: Union[float, int, LabelledCollection] = None):
""" """
Trains a ACC quantifier Trains a ACC quantifier.
:param data: the training set :param data: the training set
:param fit_learner: set to False to bypass the training (the learner is assumed to be already fit) :param fit_learner: set to False to bypass the training (the learner is assumed to be already fit)
:param val_split: either a float in (0,1) indicating the proportion of training instances to use for :param val_split: either a float in (0,1) indicating the proportion of training instances to use for
validation (e.g., 0.3 for using 30% of the training set as validation data), or a LabelledCollection validation (e.g., 0.3 for using 30% of the training set as validation data), or a LabelledCollection
indicating the validation set itself, or an int indicating the number k of folds to be used in kFCV indicating the validation set itself, or an int indicating the number `k` of folds to be used in `k`-fold
to estimate the parameters cross validation to estimate the parameters
:return: self :return: self
""" """
if val_split is None: if val_split is None:
@ -205,7 +293,7 @@ class ACC(AggregativeQuantifier):
pbar.set_description(f'{self.__class__.__name__} fitting fold {k}') pbar.set_description(f'{self.__class__.__name__} fitting fold {k}')
training = data.sampling_from_index(training_idx) training = data.sampling_from_index(training_idx)
validation = data.sampling_from_index(validation_idx) validation = data.sampling_from_index(validation_idx)
learner, val_data = training_helper(self.learner, training, fit_learner, val_split=validation) learner, val_data = _training_helper(self.learner, training, fit_learner, val_split=validation)
y_.append(learner.predict(val_data.instances)) y_.append(learner.predict(val_data.instances))
y.append(val_data.labels) y.append(val_data.labels)
@ -214,10 +302,10 @@ class ACC(AggregativeQuantifier):
class_count = data.counts() class_count = data.counts()
# fit the learner on all data # fit the learner on all data
self.learner, _ = training_helper(self.learner, data, fit_learner, val_split=None) self.learner, _ = _training_helper(self.learner, data, fit_learner, val_split=None)
else: else:
self.learner, val_data = training_helper(self.learner, data, fit_learner, val_split=val_split) self.learner, val_data = _training_helper(self.learner, data, fit_learner, val_split=val_split)
y_ = self.learner.predict(val_data.instances) y_ = self.learner.predict(val_data.instances)
y = val_data.labels y = val_data.labels
class_count = val_data.counts() class_count = val_data.counts()
@ -239,7 +327,15 @@ class ACC(AggregativeQuantifier):
@classmethod @classmethod
def solve_adjustment(cls, PteCondEstim, prevs_estim): def solve_adjustment(cls, PteCondEstim, prevs_estim):
# solve for the linear system Ax = B with A=PteCondEstim and B = prevs_estim """
Solves the system linear system :math:`Ax = B` with :math:`A` = `PteCondEstim` and :math:`B` = `prevs_estim`
:param PteCondEstim: a `np.ndarray` of shape `(n_classes,n_classes,)` with entry `(i,j)` being the estimate
of :math:`P(y_i|y_j)`, that is, the probability that an instance that belongs to :math:`y_j` ends up being
classified as belonging to :math:`y_i`
:param prevs_estim: a `np.ndarray` of shape `(n_classes,)` with the class prevalence estimates
:return: an adjusted `np.ndarray` of shape `(n_classes,)` with the corrected class prevalence estimates
"""
A = PteCondEstim A = PteCondEstim
B = prevs_estim B = prevs_estim
try: try:
@ -252,11 +348,18 @@ class ACC(AggregativeQuantifier):
class PCC(AggregativeProbabilisticQuantifier): class PCC(AggregativeProbabilisticQuantifier):
"""
`Probabilistic Classify & Count <https://ieeexplore.ieee.org/abstract/document/5694031>`_,
the probabilistic variant of CC that relies on the posterior probabilities returned by a probabilistic classifier.
:param learner: a sklearn's Estimator that generates a classifier
"""
def __init__(self, learner: BaseEstimator): def __init__(self, learner: BaseEstimator):
self.learner = learner self.learner = learner
def fit(self, data: LabelledCollection, fit_learner=True): def fit(self, data: LabelledCollection, fit_learner=True):
self.learner, _ = training_helper(self.learner, data, fit_learner, ensure_probabilistic=True) self.learner, _ = _training_helper(self.learner, data, fit_learner, ensure_probabilistic=True)
return self return self
def aggregate(self, classif_posteriors): def aggregate(self, classif_posteriors):
@ -264,6 +367,18 @@ class PCC(AggregativeProbabilisticQuantifier):
class PACC(AggregativeProbabilisticQuantifier): class PACC(AggregativeProbabilisticQuantifier):
"""
`Probabilistic Adjusted Classify & Count <https://ieeexplore.ieee.org/abstract/document/5694031>`_,
the probabilistic variant of ACC that relies on the posterior probabilities returned by a probabilistic classifier.
:param learner: a sklearn's Estimator that generates a classifier
:param val_split: indicates the proportion of data to be used as a stratified held-out validation set in which the
misclassification rates are to be estimated.
This parameter can be indicated as a real value (between 0 and 1, default 0.4), representing a proportion of
validation data, or as an integer, indicating that the misclassification rates should be estimated via
`k`-fold cross validation (this integer stands for the number of folds `k`), or as a
:class:`quapy.data.base.LabelledCollection` (the split itself).
"""
def __init__(self, learner: BaseEstimator, val_split=0.4): def __init__(self, learner: BaseEstimator, val_split=0.4):
self.learner = learner self.learner = learner
@ -271,7 +386,8 @@ class PACC(AggregativeProbabilisticQuantifier):
def fit(self, data: LabelledCollection, fit_learner=True, val_split: Union[float, int, LabelledCollection] = None): def fit(self, data: LabelledCollection, fit_learner=True, val_split: Union[float, int, LabelledCollection] = None):
""" """
Trains a PACC quantifier Trains a PACC quantifier.
:param data: the training set :param data: the training set
:param fit_learner: set to False to bypass the training (the learner is assumed to be already fit) :param fit_learner: set to False to bypass the training (the learner is assumed to be already fit)
:param val_split: either a float in (0,1) indicating the proportion of training instances to use for :param val_split: either a float in (0,1) indicating the proportion of training instances to use for
@ -294,7 +410,7 @@ class PACC(AggregativeProbabilisticQuantifier):
pbar.set_description(f'{self.__class__.__name__} fitting fold {k}') pbar.set_description(f'{self.__class__.__name__} fitting fold {k}')
training = data.sampling_from_index(training_idx) training = data.sampling_from_index(training_idx)
validation = data.sampling_from_index(validation_idx) validation = data.sampling_from_index(validation_idx)
learner, val_data = training_helper( learner, val_data = _training_helper(
self.learner, training, fit_learner, ensure_probabilistic=True, val_split=validation) self.learner, training, fit_learner, ensure_probabilistic=True, val_split=validation)
y_.append(learner.predict_proba(val_data.instances)) y_.append(learner.predict_proba(val_data.instances))
y.append(val_data.labels) y.append(val_data.labels)
@ -303,12 +419,12 @@ class PACC(AggregativeProbabilisticQuantifier):
y_ = np.vstack(y_) y_ = np.vstack(y_)
# fit the learner on all data # fit the learner on all data
self.learner, _ = training_helper(self.learner, data, fit_learner, ensure_probabilistic=True, self.learner, _ = _training_helper(self.learner, data, fit_learner, ensure_probabilistic=True,
val_split=None) val_split=None)
classes = data.classes_ classes = data.classes_
else: else:
self.learner, val_data = training_helper( self.learner, val_data = _training_helper(
self.learner, data, fit_learner, ensure_probabilistic=True, val_split=val_split) self.learner, data, fit_learner, ensure_probabilistic=True, val_split=val_split)
y_ = self.learner.predict_proba(val_data.instances) y_ = self.learner.predict_proba(val_data.instances)
y = val_data.labels y = val_data.labels
@ -337,10 +453,13 @@ class PACC(AggregativeProbabilisticQuantifier):
class EMQ(AggregativeProbabilisticQuantifier): class EMQ(AggregativeProbabilisticQuantifier):
""" """
The method is described in: `Expectation Maximization for Quantification <https://ieeexplore.ieee.org/abstract/document/6789744>`_ (EMQ),
Saerens, M., Latinne, P., and Decaestecker, C. (2002). aka `Saerens-Latinne-Decaestecker` (SLD) algorithm.
Adjusting the outputs of a classifier to new a priori probabilities: A simple procedure. EMQ consists of using the well-known `Expectation Maximization algorithm` to iteratively update the posterior
Neural Computation, 14(1): 2141. probabilities generated by a probabilistic classifier and the class prevalence estimates obtained via
maximum-likelihood estimation, in a mutually recursive way, until convergence.
:param learner: a sklearn's Estimator that generates a classifier
""" """
MAX_ITER = 1000 MAX_ITER = 1000
@ -350,7 +469,7 @@ class EMQ(AggregativeProbabilisticQuantifier):
self.learner = learner self.learner = learner
def fit(self, data: LabelledCollection, fit_learner=True): def fit(self, data: LabelledCollection, fit_learner=True):
self.learner, _ = training_helper(self.learner, data, fit_learner, ensure_probabilistic=True) self.learner, _ = _training_helper(self.learner, data, fit_learner, ensure_probabilistic=True)
self.train_prevalence = F.prevalence_from_labels(data.labels, self.classes_) self.train_prevalence = F.prevalence_from_labels(data.labels, self.classes_)
return self return self
@ -365,6 +484,17 @@ class EMQ(AggregativeProbabilisticQuantifier):
@classmethod @classmethod
def EM(cls, tr_prev, posterior_probabilities, epsilon=EPSILON): def EM(cls, tr_prev, posterior_probabilities, epsilon=EPSILON):
"""
Computes the `Expectation Maximization` routine.
:param tr_prev: array-like, the training prevalence
:param posterior_probabilities: `np.ndarray` of shape `(n_instances, n_classes,)` with the
posterior probabilities
:param epsilon: float, the threshold different between two consecutive iterations
to reach before stopping the loop
:return: a tuple with the estimated prevalence values (shape `(n_classes,)`) and
the corrected posterior probabilities (shape `(n_instances, n_classes,)`)
"""
Px = posterior_probabilities Px = posterior_probabilities
Ptr = np.copy(tr_prev) Ptr = np.copy(tr_prev)
qs = np.copy(Ptr) # qs (the running estimate) is initialized as the training prevalence qs = np.copy(Ptr) # qs (the running estimate) is initialized as the training prevalence
@ -393,9 +523,17 @@ class EMQ(AggregativeProbabilisticQuantifier):
class HDy(AggregativeProbabilisticQuantifier, BinaryQuantifier): class HDy(AggregativeProbabilisticQuantifier, BinaryQuantifier):
""" """
Implementation of the method based on the Hellinger Distance y (HDy) proposed by `Hellinger Distance y <https://www.sciencedirect.com/science/article/pii/S0020025512004069>`_ (HDy).
González-Castro, V., Alaiz-Rodrı́guez, R., and Alegre, E. (2013). Class distribution HDy is a probabilistic method for training binary quantifiers, that models quantification as the problem of
estimation based on the Hellinger distance. Information Sciences, 218:146164. minimizing the divergence (in terms of the Hellinger Distance) between two cumulative distributions of posterior
probabilities returned by the classifier. One of the distributions is generated from the unlabelled examples and
the other is generated from a validation set. This latter distribution is defined as a mixture of the
class-conditional distributions of the posterior probabilities returned for the positive and negative validation
examples, respectively. The parameters of the mixture thus represent the estimates of the class prevalence values.
:param learner: a sklearn's Estimator that generates a binary classifier
:param val_split: a float in range (0,1) indicating the proportion of data to be used as a stratified held-out
validation distribution, or a :class:`quapy.data.base.LabelledCollection` (the split itself).
""" """
def __init__(self, learner: BaseEstimator, val_split=0.4): def __init__(self, learner: BaseEstimator, val_split=0.4):
@ -404,19 +542,20 @@ class HDy(AggregativeProbabilisticQuantifier, BinaryQuantifier):
def fit(self, data: LabelledCollection, fit_learner=True, val_split: Union[float, LabelledCollection] = None): def fit(self, data: LabelledCollection, fit_learner=True, val_split: Union[float, LabelledCollection] = None):
""" """
Trains a HDy quantifier Trains a HDy quantifier.
:param data: the training set :param data: the training set
:param fit_learner: set to False to bypass the training (the learner is assumed to be already fit) :param fit_learner: set to False to bypass the training (the learner is assumed to be already fit)
:param val_split: either a float in (0,1) indicating the proportion of training instances to use for :param val_split: either a float in (0,1) indicating the proportion of training instances to use for
validation (e.g., 0.3 for using 30% of the training set as validation data), or a LabelledCollection validation (e.g., 0.3 for using 30% of the training set as validation data), or a
indicating the validation set itself :class:`quapy.data.base.LabelledCollection` indicating the validation set itself
:return: self :return: self
""" """
if val_split is None: if val_split is None:
val_split = self.val_split val_split = self.val_split
self._check_binary(data, self.__class__.__name__) self._check_binary(data, self.__class__.__name__)
self.learner, validation = training_helper( self.learner, validation = _training_helper(
self.learner, data, fit_learner, ensure_probabilistic=True, val_split=val_split) self.learner, data, fit_learner, ensure_probabilistic=True, val_split=val_split)
Px = self.posterior_probabilities(validation.instances)[:, 1] # takes only the P(y=+1|x) Px = self.posterior_probabilities(validation.instances)[:, 1] # takes only the P(y=+1|x)
self.Pxy1 = Px[validation.labels == self.learner.classes_[1]] self.Pxy1 = Px[validation.labels == self.learner.classes_[1]]
@ -459,6 +598,19 @@ class HDy(AggregativeProbabilisticQuantifier, BinaryQuantifier):
class ELM(AggregativeQuantifier, BinaryQuantifier): class ELM(AggregativeQuantifier, BinaryQuantifier):
"""
Class of Explicit Loss Minimization (ELM) quantifiers.
Quantifiers based on ELM represent a family of methods based on structured output learning;
these quantifiers rely on classifiers that have been optimized using a quantification-oriented loss
measure. This implementation relies on
`Joachims SVM perf <https://www.cs.cornell.edu/people/tj/svm_light/svm_perf.html>`_ structured output
learning algorithm, which has to be installed and patched for the purpose (see this
`script <https://github.com/HLT-ISTI/QuaPy/blob/master/prepare_svmperf.sh>`_).
:param svmperf_base: path to the folder containing the binary files of `SVM perf`
:param loss: the loss to optimize (see :attr:`quapy.classification.svmperf.SVMperf.valid_losses`)
:param kwargs: rest of SVM perf's parameters
"""
def __init__(self, svmperf_base=None, loss='01', **kwargs): def __init__(self, svmperf_base=None, loss='01', **kwargs):
self.svmperf_base = svmperf_base if svmperf_base is not None else qp.environ['SVMPERF_HOME'] self.svmperf_base = svmperf_base if svmperf_base is not None else qp.environ['SVMPERF_HOME']
@ -481,9 +633,15 @@ class ELM(AggregativeQuantifier, BinaryQuantifier):
class SVMQ(ELM): class SVMQ(ELM):
""" """
Barranquero, J., Díez, J., and del Coz, J. J. (2015). SVM(Q), which attempts to minimize the `Q` loss combining a classification-oriented loss and a
Quantification-oriented learning based on reliable classifiers. quantification-oriented loss, as proposed by
Pattern Recognition, 48(2):591604. `Barranquero et al. 2015 <https://www.sciencedirect.com/science/article/pii/S003132031400291X>`_.
Equivalent to:
>>> ELM(svmperf_base, loss='q', **kwargs)
:param svmperf_base: path to the folder containing the binary files of `SVM perf`
:param kwargs: rest of SVM perf's parameters
""" """
def __init__(self, svmperf_base=None, **kwargs): def __init__(self, svmperf_base=None, **kwargs):
@ -492,9 +650,14 @@ class SVMQ(ELM):
class SVMKLD(ELM): class SVMKLD(ELM):
""" """
Esuli, A. and Sebastiani, F. (2015). SVM(KLD), which attempts to minimize the Kullback-Leibler Divergence as proposed by
Optimizing text quantifiers for multivariate loss functions. `Esuli et al. 2015 <https://dl.acm.org/doi/abs/10.1145/2700406>`_.
ACM Transactions on Knowledge Discovery and Data, 9(4):Article 27. Equivalent to:
>>> ELM(svmperf_base, loss='kld', **kwargs)
:param svmperf_base: path to the folder containing the binary files of `SVM perf`
:param kwargs: rest of SVM perf's parameters
""" """
def __init__(self, svmperf_base=None, **kwargs): def __init__(self, svmperf_base=None, **kwargs):
@ -503,9 +666,15 @@ class SVMKLD(ELM):
class SVMNKLD(ELM): class SVMNKLD(ELM):
""" """
Esuli, A. and Sebastiani, F. (2015). SVM(NKLD), which attempts to minimize a version of the the Kullback-Leibler Divergence normalized
Optimizing text quantifiers for multivariate loss functions. via the logistic function, as proposed by
ACM Transactions on Knowledge Discovery and Data, 9(4):Article 27. `Esuli et al. 2015 <https://dl.acm.org/doi/abs/10.1145/2700406>`_.
Equivalent to:
>>> ELM(svmperf_base, loss='nkld', **kwargs)
:param svmperf_base: path to the folder containing the binary files of `SVM perf`
:param kwargs: rest of SVM perf's parameters
""" """
def __init__(self, svmperf_base=None, **kwargs): def __init__(self, svmperf_base=None, **kwargs):
@ -513,25 +682,60 @@ class SVMNKLD(ELM):
class SVMAE(ELM): class SVMAE(ELM):
"""
SVM(AE), which attempts to minimize Absolute Error as first used by
`Moreo and Sebastiani, 2021 <https://arxiv.org/abs/2011.02552>`_.
Equivalent to:
>>> ELM(svmperf_base, loss='mae', **kwargs)
:param svmperf_base: path to the folder containing the binary files of `SVM perf`
:param kwargs: rest of SVM perf's parameters
"""
def __init__(self, svmperf_base=None, **kwargs): def __init__(self, svmperf_base=None, **kwargs):
super(SVMAE, self).__init__(svmperf_base, loss='mae', **kwargs) super(SVMAE, self).__init__(svmperf_base, loss='mae', **kwargs)
class SVMRAE(ELM): class SVMRAE(ELM):
"""
SVM(RAE), which attempts to minimize Relative Absolute Error as first used by
`Moreo and Sebastiani, 2021 <https://arxiv.org/abs/2011.02552>`_.
Equivalent to:
>>> ELM(svmperf_base, loss='mrae', **kwargs)
:param svmperf_base: path to the folder containing the binary files of `SVM perf`
:param kwargs: rest of SVM perf's parameters
"""
def __init__(self, svmperf_base=None, **kwargs): def __init__(self, svmperf_base=None, **kwargs):
super(SVMRAE, self).__init__(svmperf_base, loss='mrae', **kwargs) super(SVMRAE, self).__init__(svmperf_base, loss='mrae', **kwargs)
class ThresholdOptimization(AggregativeQuantifier, BinaryQuantifier): class ThresholdOptimization(AggregativeQuantifier, BinaryQuantifier):
"""
Abstract class of Threshold Optimization variants for :class:`ACC` as proposed by
`Forman 2006 <https://dl.acm.org/doi/abs/10.1145/1150402.1150423>`_ and
`Forman 2008 <https://link.springer.com/article/10.1007/s10618-008-0097-y>`_.
The goal is to bring improved stability to the denominator of the adjustment.
The different variants are based on different heuristics for choosing a decision threshold
that would allow for more true positives and many more false positives, on the grounds this
would deliver larger denominators.
:param learner: a sklearn's Estimator that generates a classifier
:param val_split: indicates the proportion of data to be used as a stratified held-out validation set in which the
misclassification rates are to be estimated.
This parameter can be indicated as a real value (between 0 and 1, default 0.4), representing a proportion of
validation data, or as an integer, indicating that the misclassification rates should be estimated via
`k`-fold cross validation (this integer stands for the number of folds `k`), or as a
:class:`quapy.data.base.LabelledCollection` (the split itself).
"""
def __init__(self, learner: BaseEstimator, val_split=0.4): def __init__(self, learner: BaseEstimator, val_split=0.4):
self.learner = learner self.learner = learner
self.val_split = val_split self.val_split = val_split
@abstractmethod
def optimize_threshold(self, y, probabilities):
...
def fit(self, data: LabelledCollection, fit_learner=True, val_split: Union[float, int, LabelledCollection] = None): def fit(self, data: LabelledCollection, fit_learner=True, val_split: Union[float, int, LabelledCollection] = None):
self._check_binary(data, "Threshold Optimization") self._check_binary(data, "Threshold Optimization")
@ -548,7 +752,7 @@ class ThresholdOptimization(AggregativeQuantifier, BinaryQuantifier):
pbar.set_description(f'{self.__class__.__name__} fitting fold {k}') pbar.set_description(f'{self.__class__.__name__} fitting fold {k}')
training = data.sampling_from_index(training_idx) training = data.sampling_from_index(training_idx)
validation = data.sampling_from_index(validation_idx) validation = data.sampling_from_index(validation_idx)
learner, val_data = training_helper(self.learner, training, fit_learner, val_split=validation) learner, val_data = _training_helper(self.learner, training, fit_learner, val_split=validation)
probabilities.append(learner.predict_proba(val_data.instances)) probabilities.append(learner.predict_proba(val_data.instances))
y.append(val_data.labels) y.append(val_data.labels)
@ -556,16 +760,16 @@ class ThresholdOptimization(AggregativeQuantifier, BinaryQuantifier):
probabilities = np.concatenate(probabilities) probabilities = np.concatenate(probabilities)
# fit the learner on all data # fit the learner on all data
self.learner, _ = training_helper(self.learner, data, fit_learner, val_split=None) self.learner, _ = _training_helper(self.learner, data, fit_learner, val_split=None)
else: else:
self.learner, val_data = training_helper(self.learner, data, fit_learner, val_split=val_split) self.learner, val_data = _training_helper(self.learner, data, fit_learner, val_split=val_split)
probabilities = self.learner.predict_proba(val_data.instances) probabilities = self.learner.predict_proba(val_data.instances)
y = val_data.labels y = val_data.labels
self.cc = CC(self.learner) self.cc = CC(self.learner)
self.tpr, self.fpr = self.optimize_threshold(y, probabilities) self.tpr, self.fpr = self._optimize_threshold(y, probabilities)
return self return self
@ -573,20 +777,32 @@ class ThresholdOptimization(AggregativeQuantifier, BinaryQuantifier):
def _condition(self, tpr, fpr) -> float: def _condition(self, tpr, fpr) -> float:
""" """
Implements the criterion according to which the threshold should be selected. Implements the criterion according to which the threshold should be selected.
This function should return a (float) score to be minimized. This function should return the (float) score to be minimized.
:param tpr: float, true positive rate
:param fpr: float, false positive rate
:return: float, a score for the given `tpr` and `fpr`
""" """
... ...
def optimize_threshold(self, y, probabilities): def _optimize_threshold(self, y, probabilities):
"""
Seeks for the best `tpr` and `fpr` according to the score obtained at different
decision thresholds. The scoring function is implemented in function `_condition`.
:param y: predicted labels for the validation set (or for the training set via `k`-fold cross validation)
:param probabilities: array-like with the posterior probabilities
:return: best `tpr` and `fpr` according to `_condition`
"""
best_candidate_threshold_score = None best_candidate_threshold_score = None
best_tpr = 0 best_tpr = 0
best_fpr = 0 best_fpr = 0
candidate_thresholds = np.unique(probabilities[:, 1]) candidate_thresholds = np.unique(probabilities[:, 1])
for candidate_threshold in candidate_thresholds: for candidate_threshold in candidate_thresholds:
y_ = [self.classes_[1] if p > candidate_threshold else self.classes_[0] for p in probabilities[:, 1]] y_ = [self.classes_[1] if p > candidate_threshold else self.classes_[0] for p in probabilities[:, 1]]
TP, FP, FN, TN = self.compute_table(y, y_) TP, FP, FN, TN = self._compute_table(y, y_)
tpr = self.compute_tpr(TP, FP) tpr = self._compute_tpr(TP, FP)
fpr = self.compute_fpr(FP, TN) fpr = self._compute_fpr(FP, TN)
condition_score = self._condition(tpr, fpr) condition_score = self._condition(tpr, fpr)
if best_candidate_threshold_score is None or condition_score < best_candidate_threshold_score: if best_candidate_threshold_score is None or condition_score < best_candidate_threshold_score:
best_candidate_threshold_score = condition_score best_candidate_threshold_score = condition_score
@ -603,25 +819,40 @@ class ThresholdOptimization(AggregativeQuantifier, BinaryQuantifier):
adjusted_prevs_estim = np.array((1 - adjusted_prevs_estim, adjusted_prevs_estim)) adjusted_prevs_estim = np.array((1 - adjusted_prevs_estim, adjusted_prevs_estim))
return adjusted_prevs_estim return adjusted_prevs_estim
def compute_table(self, y, y_): def _compute_table(self, y, y_):
TP = np.logical_and(y == y_, y == self.classes_[1]).sum() TP = np.logical_and(y == y_, y == self.classes_[1]).sum()
FP = np.logical_and(y != y_, y == self.classes_[0]).sum() FP = np.logical_and(y != y_, y == self.classes_[0]).sum()
FN = np.logical_and(y != y_, y == self.classes_[1]).sum() FN = np.logical_and(y != y_, y == self.classes_[1]).sum()
TN = np.logical_and(y == y_, y == self.classes_[0]).sum() TN = np.logical_and(y == y_, y == self.classes_[0]).sum()
return TP, FP, FN, TN return TP, FP, FN, TN
def compute_tpr(self, TP, FP): def _compute_tpr(self, TP, FP):
if TP + FP == 0: if TP + FP == 0:
return 0 return 0
return TP / (TP + FP) return TP / (TP + FP)
def compute_fpr(self, FP, TN): def _compute_fpr(self, FP, TN):
if FP + TN == 0: if FP + TN == 0:
return 0 return 0
return FP / (FP + TN) return FP / (FP + TN)
class T50(ThresholdOptimization): class T50(ThresholdOptimization):
"""
Threshold Optimization variant for :class:`ACC` as proposed by
`Forman 2006 <https://dl.acm.org/doi/abs/10.1145/1150402.1150423>`_ and
`Forman 2008 <https://link.springer.com/article/10.1007/s10618-008-0097-y>`_ that looks
for the threshold that makes `tpr` cosest to 0.5.
The goal is to bring improved stability to the denominator of the adjustment.
:param learner: a sklearn's Estimator that generates a classifier
:param val_split: indicates the proportion of data to be used as a stratified held-out validation set in which the
misclassification rates are to be estimated.
This parameter can be indicated as a real value (between 0 and 1, default 0.4), representing a proportion of
validation data, or as an integer, indicating that the misclassification rates should be estimated via
`k`-fold cross validation (this integer stands for the number of folds `k`), or as a
:class:`quapy.data.base.LabelledCollection` (the split itself).
"""
def __init__(self, learner: BaseEstimator, val_split=0.4): def __init__(self, learner: BaseEstimator, val_split=0.4):
super().__init__(learner, val_split) super().__init__(learner, val_split)
@ -631,6 +862,21 @@ class T50(ThresholdOptimization):
class MAX(ThresholdOptimization): class MAX(ThresholdOptimization):
"""
Threshold Optimization variant for :class:`ACC` as proposed by
`Forman 2006 <https://dl.acm.org/doi/abs/10.1145/1150402.1150423>`_ and
`Forman 2008 <https://link.springer.com/article/10.1007/s10618-008-0097-y>`_ that looks
for the threshold that maximizes `tpr-fpr`.
The goal is to bring improved stability to the denominator of the adjustment.
:param learner: a sklearn's Estimator that generates a classifier
:param val_split: indicates the proportion of data to be used as a stratified held-out validation set in which the
misclassification rates are to be estimated.
This parameter can be indicated as a real value (between 0 and 1, default 0.4), representing a proportion of
validation data, or as an integer, indicating that the misclassification rates should be estimated via
`k`-fold cross validation (this integer stands for the number of folds `k`), or as a
:class:`quapy.data.base.LabelledCollection` (the split itself).
"""
def __init__(self, learner: BaseEstimator, val_split=0.4): def __init__(self, learner: BaseEstimator, val_split=0.4):
super().__init__(learner, val_split) super().__init__(learner, val_split)
@ -641,6 +887,21 @@ class MAX(ThresholdOptimization):
class X(ThresholdOptimization): class X(ThresholdOptimization):
"""
Threshold Optimization variant for :class:`ACC` as proposed by
`Forman 2006 <https://dl.acm.org/doi/abs/10.1145/1150402.1150423>`_ and
`Forman 2008 <https://link.springer.com/article/10.1007/s10618-008-0097-y>`_ that looks
for the threshold that yields `tpr=1-fpr`.
The goal is to bring improved stability to the denominator of the adjustment.
:param learner: a sklearn's Estimator that generates a classifier
:param val_split: indicates the proportion of data to be used as a stratified held-out validation set in which the
misclassification rates are to be estimated.
This parameter can be indicated as a real value (between 0 and 1, default 0.4), representing a proportion of
validation data, or as an integer, indicating that the misclassification rates should be estimated via
`k`-fold cross validation (this integer stands for the number of folds `k`), or as a
:class:`quapy.data.base.LabelledCollection` (the split itself).
"""
def __init__(self, learner: BaseEstimator, val_split=0.4): def __init__(self, learner: BaseEstimator, val_split=0.4):
super().__init__(learner, val_split) super().__init__(learner, val_split)
@ -650,41 +911,70 @@ class X(ThresholdOptimization):
class MS(ThresholdOptimization): class MS(ThresholdOptimization):
"""
Median Sweep. Threshold Optimization variant for :class:`ACC` as proposed by
`Forman 2006 <https://dl.acm.org/doi/abs/10.1145/1150402.1150423>`_ and
`Forman 2008 <https://link.springer.com/article/10.1007/s10618-008-0097-y>`_ that generates
class prevalence estimates for all decision thresholds and returns the median of them all.
The goal is to bring improved stability to the denominator of the adjustment.
:param learner: a sklearn's Estimator that generates a classifier
:param val_split: indicates the proportion of data to be used as a stratified held-out validation set in which the
misclassification rates are to be estimated.
This parameter can be indicated as a real value (between 0 and 1, default 0.4), representing a proportion of
validation data, or as an integer, indicating that the misclassification rates should be estimated via
`k`-fold cross validation (this integer stands for the number of folds `k`), or as a
:class:`quapy.data.base.LabelledCollection` (the split itself).
"""
def __init__(self, learner: BaseEstimator, val_split=0.4): def __init__(self, learner: BaseEstimator, val_split=0.4):
super().__init__(learner, val_split) super().__init__(learner, val_split)
def _condition(self, tpr, fpr) -> float: def _condition(self, tpr, fpr) -> float:
pass pass
def optimize_threshold(self, y, probabilities): def _optimize_threshold(self, y, probabilities):
tprs = [] tprs = []
fprs = [] fprs = []
candidate_thresholds = np.unique(probabilities[:, 1]) candidate_thresholds = np.unique(probabilities[:, 1])
for candidate_threshold in candidate_thresholds: for candidate_threshold in candidate_thresholds:
y_ = [self.classes_[1] if p > candidate_threshold else self.classes_[0] for p in probabilities[:, 1]] y_ = [self.classes_[1] if p > candidate_threshold else self.classes_[0] for p in probabilities[:, 1]]
TP, FP, FN, TN = self.compute_table(y, y_) TP, FP, FN, TN = self._compute_table(y, y_)
tpr = self.compute_tpr(TP, FP) tpr = self._compute_tpr(TP, FP)
fpr = self.compute_fpr(FP, TN) fpr = self._compute_fpr(FP, TN)
tprs.append(tpr) tprs.append(tpr)
fprs.append(fpr) fprs.append(fpr)
return np.median(tprs), np.median(fprs) return np.median(tprs), np.median(fprs)
class MS2(MS): class MS2(MS):
"""
Median Sweep 2. Threshold Optimization variant for :class:`ACC` as proposed by
`Forman 2006 <https://dl.acm.org/doi/abs/10.1145/1150402.1150423>`_ and
`Forman 2008 <https://link.springer.com/article/10.1007/s10618-008-0097-y>`_ that generates
class prevalence estimates for all decision thresholds and returns the median of for cases in
which `tpr-fpr>0.25`
The goal is to bring improved stability to the denominator of the adjustment.
:param learner: a sklearn's Estimator that generates a classifier
:param val_split: indicates the proportion of data to be used as a stratified held-out validation set in which the
misclassification rates are to be estimated.
This parameter can be indicated as a real value (between 0 and 1, default 0.4), representing a proportion of
validation data, or as an integer, indicating that the misclassification rates should be estimated via
`k`-fold cross validation (this integer stands for the number of folds `k`), or as a
:class:`quapy.data.base.LabelledCollection` (the split itself).
"""
def __init__(self, learner: BaseEstimator, val_split=0.4): def __init__(self, learner: BaseEstimator, val_split=0.4):
super().__init__(learner, val_split) super().__init__(learner, val_split)
def optimize_threshold(self, y, probabilities): def _optimize_threshold(self, y, probabilities):
tprs = [0, 1] tprs = [0, 1]
fprs = [0, 1] fprs = [0, 1]
candidate_thresholds = np.unique(probabilities[:, 1]) candidate_thresholds = np.unique(probabilities[:, 1])
for candidate_threshold in candidate_thresholds: for candidate_threshold in candidate_thresholds:
y_ = [self.classes_[1] if p > candidate_threshold else self.classes_[0] for p in probabilities[:, 1]] y_ = [self.classes_[1] if p > candidate_threshold else self.classes_[0] for p in probabilities[:, 1]]
TP, FP, FN, TN = self.compute_table(y, y_) TP, FP, FN, TN = self._compute_table(y, y_)
tpr = self.compute_tpr(TP, FP) tpr = self._compute_tpr(TP, FP)
fpr = self.compute_fpr(FP, TN) fpr = self._compute_fpr(FP, TN)
if (tpr - fpr) > 0.25: if (tpr - fpr) > 0.25:
tprs.append(tpr) tprs.append(tpr)
fprs.append(fpr) fprs.append(fpr)
@ -696,6 +986,7 @@ AdjustedClassifyAndCount = ACC
ProbabilisticClassifyAndCount = PCC ProbabilisticClassifyAndCount = PCC
ProbabilisticAdjustedClassifyAndCount = PACC ProbabilisticAdjustedClassifyAndCount = PACC
ExpectationMaximizationQuantifier = EMQ ExpectationMaximizationQuantifier = EMQ
SLD = EMQ
HellingerDistanceY = HDy HellingerDistanceY = HDy
ExplicitLossMinimisation = ELM ExplicitLossMinimisation = ELM
MedianSweep = MS MedianSweep = MS
@ -704,11 +995,14 @@ MedianSweep2 = MS2
class OneVsAll(AggregativeQuantifier): class OneVsAll(AggregativeQuantifier):
""" """
Allows any binary quantifier to perform quantification on single-label datasets. The method maintains one binary Allows any binary quantifier to perform quantification on single-label datasets.
quantifier for each class, and then l1-normalizes the outputs so that the class prevelences sum up to 1. The method maintains one binary quantifier for each class, and then l1-normalizes the outputs so that the
This variant was used, along with the ExplicitLossMinimization quantifier in class prevelences sum up to 1.
Gao, W., Sebastiani, F.: From classification to quantification in tweet sentiment analysis. This variant was used, along with the :class:`EMQ` quantifier, in
Social Network Analysis and Mining 6(19), 122 (2016) `Gao and Sebastiani, 2016 <https://link.springer.com/content/pdf/10.1007/s13278-016-0327-z.pdf>`_.
:param learner: a sklearn's Estimator that generates a binary classifier
:param n_jobs: number of parallel workers
""" """
def __init__(self, binary_quantifier, n_jobs=-1): def __init__(self, binary_quantifier, n_jobs=-1):
@ -727,18 +1021,30 @@ class OneVsAll(AggregativeQuantifier):
return self return self
def classify(self, instances): def classify(self, instances):
# returns a matrix of shape (n,m) with n the number of instances and m the number of classes. The entry """
# (i,j) is a binary value indicating whether instance i belongs to class j. The binary classifications are Returns a matrix of shape `(n,m,)` with `n` the number of instances and `m` the number of classes. The entry
# independent of each other, meaning that an instance can end up be attributed to 0, 1, or more classes. `(i,j)` is a binary value indicating whether instance `i `belongs to class `j`. The binary classifications are
independent of each other, meaning that an instance can end up be attributed to 0, 1, or more classes.
:param instances: array-like
:return: `np.ndarray`
"""
classif_predictions_bin = self.__parallel(self._delayed_binary_classification, instances) classif_predictions_bin = self.__parallel(self._delayed_binary_classification, instances)
return classif_predictions_bin.T return classif_predictions_bin.T
def posterior_probabilities(self, instances): def posterior_probabilities(self, instances):
# returns a matrix of shape (n,m,2) with n the number of instances and m the number of classes. The entry """
# (i,j,1) (resp. (i,j,0)) is a value in [0,1] indicating the posterior probability that instance i belongs Returns a matrix of shape `(n,m,2)` with `n` the number of instances and `m` the number of classes. The entry
# (resp. does not belong) to class j. `(i,j,1)` (resp. `(i,j,0)`) is a value in [0,1] indicating the posterior probability that instance `i` belongs
# The posterior probabilities are independent of each other, meaning that, in general, they do not sum (resp. does not belong) to class `j`.
# up to one. The posterior probabilities are independent of each other, meaning that, in general, they do not sum
up to one.
:param instances: array-like
:return: `np.ndarray`
"""
if not self.binary_quantifier.probabilistic: if not self.binary_quantifier.probabilistic:
raise NotImplementedError(f'{self.__class__.__name__} does not implement posterior_probabilities because ' raise NotImplementedError(f'{self.__class__.__name__} does not implement posterior_probabilities because '
f'the base quantifier {self.binary_quantifier.__class__.__name__} is not ' f'the base quantifier {self.binary_quantifier.__class__.__name__} is not '
@ -800,8 +1106,19 @@ class OneVsAll(AggregativeQuantifier):
@property @property
def binary(self): def binary(self):
"""
Informs that the classifier is not binary
:return: False
"""
return False return False
@property @property
def probabilistic(self): def probabilistic(self):
"""
Indicates if the classifier is probabilistic or not (depending on the nature of the base classifier).
:return: boolean
"""
return self.binary_quantifier.probabilistic return self.binary_quantifier.probabilistic

View File

@ -6,39 +6,107 @@ from quapy.data import LabelledCollection
# Base Quantifier abstract class # Base Quantifier abstract class
# ------------------------------------ # ------------------------------------
class BaseQuantifier(metaclass=ABCMeta): class BaseQuantifier(metaclass=ABCMeta):
"""
Abstract Quantifier. A quantifier is defined as an object of a class that implements the method :meth:`fit` on
:class:`quapy.data.base.LabelledCollection`, the method :meth:`quantify`, and the :meth:`set_params` and
:meth:`get_params` for model selection (see :meth:`quapy.model_selection.GridSearchQ`)
"""
@abstractmethod @abstractmethod
def fit(self, data: LabelledCollection): ... def fit(self, data: LabelledCollection):
"""
Trains a quantifier.
:param data: a :class:`quapy.data.base.LabelledCollection` consisting of the training data
:return: self
"""
...
@abstractmethod @abstractmethod
def quantify(self, instances): ... def quantify(self, instances):
"""
Generate class prevalence estimates for the sample's instances
:param instances: array-like
:return: `np.ndarray` of shape `(self.n_classes_,)` with class prevalence estimates.
"""
...
@abstractmethod @abstractmethod
def set_params(self, **parameters): ... def set_params(self, **parameters):
"""
Set the parameters of the quantifier.
:param parameters: dictionary of param-value pairs
"""
...
@abstractmethod @abstractmethod
def get_params(self, deep=True): ... def get_params(self, deep=True):
"""
Return the current parameters of the quantifier.
:param deep: for compatibility with sklearn
:return: a dictionary of param-value pairs
"""
...
@property @property
@abstractmethod @abstractmethod
def classes_(self): ... def classes_(self):
"""
Class labels, in the same order in which class prevalence values are to be computed.
:return: array-like
"""
...
@property
def n_classes(self):
"""
Returns the number of classes
:return: integer
"""
return len(self.classes_)
# these methods allows meta-learners to reimplement the decision based on their constituents, and not # these methods allows meta-learners to reimplement the decision based on their constituents, and not
# based on class structure # based on class structure
@property @property
def binary(self): def binary(self):
"""
Indicates whether the quantifier is binary or not.
:return: False (to be overridden)
"""
return False return False
@property @property
def aggregative(self): def aggregative(self):
"""
Indicates whether the quantifier is of type aggregative or not
:return: False (to be overridden)
"""
return False return False
@property @property
def probabilistic(self): def probabilistic(self):
"""
Indicates whether the quantifier is of type probabilistic or not
:return: False (to be overridden)
"""
return False return False
class BinaryQuantifier(BaseQuantifier): class BinaryQuantifier(BaseQuantifier):
"""
Abstract class of binary quantifiers, i.e., quantifiers estimating class prevalence values for only two classes
(typically, to be interpreted as one class and its complement).
"""
def _check_binary(self, data: LabelledCollection, quantifier_name): def _check_binary(self, data: LabelledCollection, quantifier_name):
assert data.binary, f'{quantifier_name} works only on problems of binary classification. ' \ assert data.binary, f'{quantifier_name} works only on problems of binary classification. ' \
@ -46,18 +114,43 @@ class BinaryQuantifier(BaseQuantifier):
@property @property
def binary(self): def binary(self):
"""
Informs that the quantifier is binary
:return: True
"""
return True return True
def isbinary(model:BaseQuantifier): def isbinary(model:BaseQuantifier):
"""
Alias for property `binary`
:param model: the model
:return: True if the model is binary, False otherwise
"""
return model.binary return model.binary
def isaggregative(model:BaseQuantifier): def isaggregative(model:BaseQuantifier):
"""
Alias for property `aggregative`
:param model: the model
:return: True if the model is aggregative, False otherwise
"""
return model.aggregative return model.aggregative
def isprobabilistic(model:BaseQuantifier): def isprobabilistic(model:BaseQuantifier):
"""
Alias for property `probabilistic`
:param model: the model
:return: True if the model is probabilistic, False otherwise
"""
return model.probabilistic return model.probabilistic

View File

@ -1,6 +1,5 @@
from copy import deepcopy from copy import deepcopy
from typing import Union from typing import Union
import numpy as np import numpy as np
from sklearn.linear_model import LogisticRegression from sklearn.linear_model import LogisticRegression
from sklearn.metrics import f1_score, make_scorer, accuracy_score from sklearn.metrics import f1_score, make_scorer, accuracy_score
@ -30,14 +29,40 @@ class Ensemble(BaseQuantifier):
VALID_POLICIES = {'ave', 'ptr', 'ds'} | qp.error.QUANTIFICATION_ERROR_NAMES VALID_POLICIES = {'ave', 'ptr', 'ds'} | qp.error.QUANTIFICATION_ERROR_NAMES
""" """
Methods from the articles: Implementation of the Ensemble methods for quantification described by
Pérez-Gállego, P., Quevedo, J. R., & del Coz, J. J. (2017). `Pérez-Gállego et al., 2017 <https://www.sciencedirect.com/science/article/pii/S1566253516300628>`_
Using ensembles for problems with characterizable changes in data distribution: A case study on quantification.
Information Fusion, 34, 87-100.
and and
Pérez-Gállego, P., Castano, A., Quevedo, J. R., & del Coz, J. J. (2019). `Pérez-Gállego et al., 2019 <https://www.sciencedirect.com/science/article/pii/S1566253517303652>`_.
Dynamic ensemble selection for quantification tasks. The policies implemented include:
Information Fusion, 45, 1-15.
- Average (`policy='ave'`): computes class prevalence estimates as the average of the estimates
returned by the base quantifiers.
- Training Prevalence (`policy='ptr'`): applies a dynamic selection to the ensembles members by retaining only
those members such that the class prevalence values in the samples they use as training set are closest to
preliminary class prevalence estimates computed as the average of the estimates of all the members. The final
estimate is recomputed by considering only the selected members.
- Distribution Similarity (`policy='ds'`): performs a dynamic selection of base members by retaining
the members trained on samples whose distribution of posterior probabilities is closest, in terms of the
Hellinger Distance, to the distribution of posterior probabilities in the test sample
- Accuracy (`policy='<valid error name>'`): performs a static selection of the ensemble members by
retaining those that minimize a quantification error measure, which is passed as an argument.
Example:
>>> model = Ensemble(quantifier=ACC(LogisticRegression()), size=30, policy='ave', n_jobs=-1)
:param quantifier: base quantification member of the ensemble
:param size: number of members
:param red_size: number of members to retain after selection (depending on the policy)
:param min_pos: minimum number of positive instances to consider a sample as valid
:param policy: the selection policy; available policies include: `ave` (default), `ptr`, `ds`, and accuracy
(which is instantiated via a valid error name, e.g., `mae`)
:param max_sample_size: maximum number of instances to consider in the samples (set to None
to indicate no limit, default)
:param val_split: a float in range (0,1) indicating the proportion of data to be used as a stratified held-out
validation split, or a :class:`quapy.data.base.LabelledCollection` (the split itself).
:param n_jobs: number of parallel workers (default 1)
:param verbose: set to True (default is False) to get some information in standard output
""" """
def __init__(self, def __init__(self,
@ -47,7 +72,7 @@ class Ensemble(BaseQuantifier):
min_pos=5, min_pos=5,
policy='ave', policy='ave',
max_sample_size=None, max_sample_size=None,
val_split=None, val_split:Union[qp.data.LabelledCollection, float]=None,
n_jobs=1, n_jobs=1,
verbose=False): verbose=False):
assert policy in Ensemble.VALID_POLICIES, \ assert policy in Ensemble.VALID_POLICIES, \
@ -65,12 +90,12 @@ class Ensemble(BaseQuantifier):
self.verbose = verbose self.verbose = verbose
self.max_sample_size = max_sample_size self.max_sample_size = max_sample_size
def sout(self, msg): def _sout(self, msg):
if self.verbose: if self.verbose:
print('[Ensemble]' + msg) print('[Ensemble]' + msg)
def fit(self, data: qp.data.LabelledCollection, val_split: Union[qp.data.LabelledCollection, float] = None): def fit(self, data: qp.data.LabelledCollection, val_split: Union[qp.data.LabelledCollection, float] = None):
self.sout('Fit') self._sout('Fit')
if self.policy == 'ds' and not data.binary: if self.policy == 'ds' and not data.binary:
raise ValueError(f'ds policy is only defined for binary quantification, but this dataset is not binary') raise ValueError(f'ds policy is only defined for binary quantification, but this dataset is not binary')
if val_split is None: if val_split is None:
@ -84,7 +109,7 @@ class Ensemble(BaseQuantifier):
posteriors = None posteriors = None
if self.policy == 'ds': if self.policy == 'ds':
# precompute the training posterior probabilities # precompute the training posterior probabilities
posteriors, self.post_proba_fn = self.ds_policy_get_posteriors(data) posteriors, self.post_proba_fn = self._ds_policy_get_posteriors(data)
is_static_policy = (self.policy in qp.error.QUANTIFICATION_ERROR_NAMES) is_static_policy = (self.policy in qp.error.QUANTIFICATION_ERROR_NAMES)
@ -99,9 +124,9 @@ class Ensemble(BaseQuantifier):
# static selection policy (the name of a quantification-oriented error function to minimize) # static selection policy (the name of a quantification-oriented error function to minimize)
if self.policy in qp.error.QUANTIFICATION_ERROR_NAMES: if self.policy in qp.error.QUANTIFICATION_ERROR_NAMES:
self.accuracy_policy(error_name=self.policy) self._accuracy_policy(error_name=self.policy)
self.sout('Fit [Done]') self._sout('Fit [Done]')
return self return self
def quantify(self, instances): def quantify(self, instances):
@ -110,23 +135,42 @@ class Ensemble(BaseQuantifier):
) )
if self.policy == 'ptr': if self.policy == 'ptr':
predictions = self.ptr_policy(predictions) predictions = self._ptr_policy(predictions)
elif self.policy == 'ds': elif self.policy == 'ds':
predictions = self.ds_policy(predictions, instances) predictions = self._ds_policy(predictions, instances)
predictions = np.mean(predictions, axis=0) predictions = np.mean(predictions, axis=0)
return F.normalize_prevalence(predictions) return F.normalize_prevalence(predictions)
def set_params(self, **parameters): def set_params(self, **parameters):
"""
This function should not be used within :class:`quapy.model_selection.GridSearchQ` (is here for compatibility
with the abstract class).
Instead, use `Ensemble(GridSearchQ(q),...)`, with `q` a Quantifier (recommended), or
`Ensemble(Q(GridSearchCV(l)))` with `Q` a quantifier class that has a learner `l` optimized for
classification (not recommended).
:param parameters: dictionary
:return: raises an Exception
"""
raise NotImplementedError(f'{self.__class__.__name__} should not be used within GridSearchQ; ' raise NotImplementedError(f'{self.__class__.__name__} should not be used within GridSearchQ; '
f'instead, use Ensemble(GridSearchQ(q),...), with q a Quantifier (recommended), ' f'instead, use Ensemble(GridSearchQ(q),...), with q a Quantifier (recommended), '
f'or Ensemble(Q(GridSearchCV(l))) with Q a quantifier class that has a learner ' f'or Ensemble(Q(GridSearchCV(l))) with Q a quantifier class that has a learner '
f'l optimized for classification (not recommended).') f'l optimized for classification (not recommended).')
def get_params(self, deep=True): def get_params(self, deep=True):
"""
This function should not be used within :class:`quapy.model_selection.GridSearchQ` (is here for compatibility
with the abstract class).
Instead, use `Ensemble(GridSearchQ(q),...)`, with `q` a Quantifier (recommended), or
`Ensemble(Q(GridSearchCV(l)))` with `Q` a quantifier class that has a learner `l` optimized for
classification (not recommended).
:return: raises an Exception
"""
raise NotImplementedError() raise NotImplementedError()
def accuracy_policy(self, error_name): def _accuracy_policy(self, error_name):
""" """
Selects the red_size best performant quantifiers in a static way (i.e., dropping all non-selected instances). Selects the red_size best performant quantifiers in a static way (i.e., dropping all non-selected instances).
For each model in the ensemble, the performance is measured in terms of _error_name_ on the quantification of For each model in the ensemble, the performance is measured in terms of _error_name_ on the quantification of
@ -141,7 +185,7 @@ class Ensemble(BaseQuantifier):
self.ensemble = _select_k(self.ensemble, order, k=self.red_size) self.ensemble = _select_k(self.ensemble, order, k=self.red_size)
def ptr_policy(self, predictions): def _ptr_policy(self, predictions):
""" """
Selects the predictions made by models that have been trained on samples with a prevalence that is most similar Selects the predictions made by models that have been trained on samples with a prevalence that is most similar
to a first approximation of the test prevalence as made by all models in the ensemble. to a first approximation of the test prevalence as made by all models in the ensemble.
@ -152,7 +196,7 @@ class Ensemble(BaseQuantifier):
order = np.argsort(ptr_differences) order = np.argsort(ptr_differences)
return _select_k(predictions, order, k=self.red_size) return _select_k(predictions, order, k=self.red_size)
def ds_policy_get_posteriors(self, data: LabelledCollection): def _ds_policy_get_posteriors(self, data: LabelledCollection):
""" """
In the original article, this procedure is not described in a sufficient level of detail. The paper only says In the original article, this procedure is not described in a sufficient level of detail. The paper only says
that the distribution of posterior probabilities from training and test examples is compared by means of the that the distribution of posterior probabilities from training and test examples is compared by means of the
@ -182,7 +226,7 @@ class Ensemble(BaseQuantifier):
return posteriors, posteriors_generator return posteriors, posteriors_generator
def ds_policy(self, predictions, test): def _ds_policy(self, predictions, test):
test_posteriors = self.post_proba_fn(test) test_posteriors = self.post_proba_fn(test)
test_distribution = get_probability_distribution(test_posteriors) test_distribution = get_probability_distribution(test_posteriors)
tr_distributions = [m[2] for m in self.ensemble] tr_distributions = [m[2] for m in self.ensemble]
@ -196,18 +240,40 @@ class Ensemble(BaseQuantifier):
@property @property
def binary(self): def binary(self):
"""
Returns a boolean indicating whether the base quantifiers are binary or not
:return: boolean
"""
return self.base_quantifier.binary return self.base_quantifier.binary
@property @property
def aggregative(self): def aggregative(self):
"""
Indicates that the quantifier is not aggregative.
:return: False
"""
return False return False
@property @property
def probabilistic(self): def probabilistic(self):
"""
Indicates that the quantifier is not probabilistic.
:return: False
"""
return False return False
def get_probability_distribution(posterior_probabilities, bins=8): def get_probability_distribution(posterior_probabilities, bins=8):
"""
Gets a histogram out of the posterior probabilities (only for the binary case).
:param posterior_probabilities: array-like of shape `(n_instances, 2,)`
:param bins: integer
:return: `np.ndarray` with the relative frequencies for each bin (for the positive class only)
"""
assert posterior_probabilities.shape[1] == 2, 'the posterior probabilities do not seem to be for a binary problem' assert posterior_probabilities.shape[1] == 2, 'the posterior probabilities do not seem to be for a binary problem'
posterior_probabilities = posterior_probabilities[:, 1] # take the positive posteriors only posterior_probabilities = posterior_probabilities[:, 1] # take the positive posteriors only
distribution, _ = np.histogram(posterior_probabilities, bins=bins, range=(0, 1), density=True) distribution, _ = np.histogram(posterior_probabilities, bins=bins, range=(0, 1), density=True)
@ -306,6 +372,23 @@ def _check_error(error):
def ensembleFactory(learner, base_quantifier_class, param_grid=None, optim=None, param_model_sel: dict = None, def ensembleFactory(learner, base_quantifier_class, param_grid=None, optim=None, param_model_sel: dict = None,
**kwargs): **kwargs):
"""
Ensemble factory. Provides a unified interface for instantiating ensembles that can be optimized (via model
selection for quantification) for a given evaluation metric using :class:`quapy.model_selection.GridSearchQ`.
If the evaluation metric is classification-oriented
(instead of quantification-oriented), then the optimization will be carried out via sklearn's
`GridSearchCV <https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.GridSearchCV.html>`_.
:param learner: sklearn's Estimator that generates a classifier
:param base_quantifier_class: a class of quantifiers
:param param_grid: a dictionary with the grid of parameters to optimize for
:param optim: a valid quantification or classification error, or a string name of it
:param param_model_sel: a dictionary containing any keyworded argument to pass to
:class:`quapy.model_selection.GridSearchQ`
:param kwargs: kwargs for the class :class:`Ensemble`
:return: an instance of :class:`Ensemble`
"""
if optim is not None: if optim is not None:
if param_grid is None: if param_grid is None:
raise ValueError(f'param_grid is None but optim was requested.') raise ValueError(f'param_grid is None but optim was requested.')
@ -316,20 +399,83 @@ def ensembleFactory(learner, base_quantifier_class, param_grid=None, optim=None,
def ECC(learner, param_grid=None, optim=None, param_mod_sel=None, **kwargs): def ECC(learner, param_grid=None, optim=None, param_mod_sel=None, **kwargs):
"""
Implements an ensemble of :class:`quapy.method.aggregative.CC` quantifiers, as used by
`Pérez-Gállego et al., 2019 <https://www.sciencedirect.com/science/article/pii/S1566253517303652>`_.
:param learner: sklearn's Estimator that generates a classifier
:param param_grid: a dictionary with the grid of parameters to optimize for
:param optim: a valid quantification or classification error, or a string name of it
:param param_model_sel: a dictionary containing any keyworded argument to pass to
:class:`quapy.model_selection.GridSearchQ`
:param kwargs: kwargs for the class :class:`Ensemble`
:return: an instance of :class:`Ensemble`
"""
return ensembleFactory(learner, CC, param_grid, optim, param_mod_sel, **kwargs) return ensembleFactory(learner, CC, param_grid, optim, param_mod_sel, **kwargs)
def EACC(learner, param_grid=None, optim=None, param_mod_sel=None, **kwargs): def EACC(learner, param_grid=None, optim=None, param_mod_sel=None, **kwargs):
"""
Implements an ensemble of :class:`quapy.method.aggregative.ACC` quantifiers, as used by
`Pérez-Gállego et al., 2019 <https://www.sciencedirect.com/science/article/pii/S1566253517303652>`_.
:param learner: sklearn's Estimator that generates a classifier
:param param_grid: a dictionary with the grid of parameters to optimize for
:param optim: a valid quantification or classification error, or a string name of it
:param param_model_sel: a dictionary containing any keyworded argument to pass to
:class:`quapy.model_selection.GridSearchQ`
:param kwargs: kwargs for the class :class:`Ensemble`
:return: an instance of :class:`Ensemble`
"""
return ensembleFactory(learner, ACC, param_grid, optim, param_mod_sel, **kwargs) return ensembleFactory(learner, ACC, param_grid, optim, param_mod_sel, **kwargs)
def EPACC(learner, param_grid=None, optim=None, param_mod_sel=None, **kwargs): def EPACC(learner, param_grid=None, optim=None, param_mod_sel=None, **kwargs):
"""
Implements an ensemble of :class:`quapy.method.aggregative.PACC` quantifiers.
:param learner: sklearn's Estimator that generates a classifier
:param param_grid: a dictionary with the grid of parameters to optimize for
:param optim: a valid quantification or classification error, or a string name of it
:param param_model_sel: a dictionary containing any keyworded argument to pass to
:class:`quapy.model_selection.GridSearchQ`
:param kwargs: kwargs for the class :class:`Ensemble`
:return: an instance of :class:`Ensemble`
"""
return ensembleFactory(learner, PACC, param_grid, optim, param_mod_sel, **kwargs) return ensembleFactory(learner, PACC, param_grid, optim, param_mod_sel, **kwargs)
def EHDy(learner, param_grid=None, optim=None, param_mod_sel=None, **kwargs): def EHDy(learner, param_grid=None, optim=None, param_mod_sel=None, **kwargs):
"""
Implements an ensemble of :class:`quapy.method.aggregative.HDy` quantifiers, as used by
`Pérez-Gállego et al., 2019 <https://www.sciencedirect.com/science/article/pii/S1566253517303652>`_.
:param learner: sklearn's Estimator that generates a classifier
:param param_grid: a dictionary with the grid of parameters to optimize for
:param optim: a valid quantification or classification error, or a string name of it
:param param_model_sel: a dictionary containing any keyworded argument to pass to
:class:`quapy.model_selection.GridSearchQ`
:param kwargs: kwargs for the class :class:`Ensemble`
:return: an instance of :class:`Ensemble`
"""
return ensembleFactory(learner, HDy, param_grid, optim, param_mod_sel, **kwargs) return ensembleFactory(learner, HDy, param_grid, optim, param_mod_sel, **kwargs)
def EEMQ(learner, param_grid=None, optim=None, param_mod_sel=None, **kwargs): def EEMQ(learner, param_grid=None, optim=None, param_mod_sel=None, **kwargs):
"""
Implements an ensemble of :class:`quapy.method.aggregative.EMQ` quantifiers.
:param learner: sklearn's Estimator that generates a classifier
:param param_grid: a dictionary with the grid of parameters to optimize for
:param optim: a valid quantification or classification error, or a string name of it
:param param_model_sel: a dictionary containing any keyworded argument to pass to
:class:`quapy.model_selection.GridSearchQ`
:param kwargs: kwargs for the class :class:`Ensemble`
:return: an instance of :class:`Ensemble`
"""
return ensembleFactory(learner, EMQ, param_grid, optim, param_mod_sel, **kwargs) return ensembleFactory(learner, EMQ, param_grid, optim, param_mod_sel, **kwargs)

View File

@ -62,6 +62,8 @@ class QuaNetTrainer(BaseQuantifier):
def fit(self, data: LabelledCollection, fit_learner=True): def fit(self, data: LabelledCollection, fit_learner=True):
""" """
Trains QuaNet.
:param data: the training data on which to train QuaNet. If fit_learner=True, the data will be split in :param data: the training data on which to train QuaNet. If fit_learner=True, the data will be split in
40/40/20 for training the classifier, training QuaNet, and validating QuaNet, respectively. If 40/40/20 for training the classifier, training QuaNet, and validating QuaNet, respectively. If
fit_learner=False, the data will be split in 66/34 for training QuaNet and validating it, respectively. fit_learner=False, the data will be split in 66/34 for training QuaNet and validating it, respectively.

View File

@ -3,24 +3,60 @@ from .base import BaseQuantifier
class MaximumLikelihoodPrevalenceEstimation(BaseQuantifier): class MaximumLikelihoodPrevalenceEstimation(BaseQuantifier):
"""
The `Maximum Likelihood Prevalence Estimation` (MLPE) method is a lazy method that assumes there is no prior
probability shift between training and test instances (put it other way, that the i.i.d. assumpion holds).
The estimation of class prevalence values for any test sample is always (i.e., irrespective of the test sample
itself) the class prevalence seen during training. This method is considered to be a lower-bound quantifier that
any quantification method should beat.
"""
def __init__(self, **kwargs): def __init__(self):
self._classes_ = None self._classes_ = None
def fit(self, data: LabelledCollection, *args): def fit(self, data: LabelledCollection):
"""
Computes the training prevalence and stores it.
:param data: the training sample
:return: self
"""
self._classes_ = data.classes_ self._classes_ = data.classes_
self.estimated_prevalence = data.prevalence() self.estimated_prevalence = data.prevalence()
return self return self
def quantify(self, documents, *args): def quantify(self, instances):
"""
Ignores the input instances and returns, as the class prevalence estimantes, the training prevalence.
:param instances: array-like (ignored)
:return: the class prevalence seen during training
"""
return self.estimated_prevalence return self.estimated_prevalence
@property @property
def classes_(self): def classes_(self):
"""
Number of classes
:return: integer
"""
return self._classes_ return self._classes_
def get_params(self): def get_params(self, deep=True):
pass """
Does nothing, since this learner has no parameters.
:param deep: for compatibility with sklearn
:return: `None`
"""
return None
def set_params(self, **parameters): def set_params(self, **parameters):
"""
Does nothing, since this learner has no parameters.
:param parameters: dictionary of param-value pairs (ignored)
"""
pass pass