Commits

Aleš Erjavec committed 4e777e3

Moved _reliability package into orangecontrib namespace.

  • Participants
  • Parent commits 21240b9

Comments (0)

Files changed (12)

-recursive-include _reliability *.png
+recursive-include orangecontrib/reliability *.svg
 recursive-include docs/rst *
 include docs/Makefile
 include COPYING

File _reliability/__init__.py

-import Orange
-
-import random
-from Orange import statc
-import math
-import warnings
-import numpy
-
-from collections import defaultdict
-from itertools import izip
-
-# Labels and final variables
-labels = ["SAvar", "SAbias", "BAGV", "CNK", "LCV", "BVCK", "Mahalanobis", "ICV"]
-
-"""
-# All the estimators calculation constants
-DO_SA = 0
-DO_BAGV = 1
-DO_CNK = 2
-DO_LCV = 3
-DO_BVCK = 4
-DO_MAHAL = 5
-"""
-
-# All the estimator method constants
-SAVAR_ABSOLUTE = 0
-SABIAS_SIGNED = 1
-SABIAS_ABSOLUTE = 2
-BAGV_ABSOLUTE = 3
-CNK_SIGNED = 4
-CNK_ABSOLUTE = 5
-LCV_ABSOLUTE = 6
-BVCK_ABSOLUTE = 7
-MAHAL_ABSOLUTE = 8
-BLENDING_ABSOLUTE = 9
-ICV_METHOD = 10
-MAHAL_TO_CENTER_ABSOLUTE = 13
-DENS_ABSOLUTE = 14
-ERR_ABSOLUTE = 15
-
-# Type of estimator constant
-SIGNED = 0
-ABSOLUTE = 1
-
-# Names of all the estimator methods
-METHOD_NAME = {0: "SAvar absolute", 1: "SAbias signed", 2: "SAbias absolute",
-               3: "BAGV absolute", 4: "CNK signed", 5: "CNK absolute",
-               6: "LCV absolute", 7: "BVCK_absolute", 8: "Mahalanobis absolute",
-               9: "BLENDING absolute", 10: "ICV", 11: "RF Variance", 12: "RF Std",
-               13: "Mahalanobis to center", 14: "Density based", 15: "Reference expected error"}
-
-select_with_repeat = Orange.core.MakeRandomIndicesMultiple()
-select_with_repeat.random_generator = Orange.misc.Random()
-
-def get_reliability_estimation_list(res, i):
-    return [result.probabilities[0].reliability_estimate[i].estimate for result in res.results], res.results[0].probabilities[0].reliability_estimate[i].signed_or_absolute, res.results[0].probabilities[0].reliability_estimate[i].method
-
-def get_prediction_error_list(res):
-    return [result.actual_class - result.classes[0] for result in res.results]
-
-def get_description_list(res, i):
-    return [result.probabilities[0].reliability_estimate[i].text_description for result in res.results]
-
-def get_pearson_r(res):
-    """
-    :param res: results of evaluation, done using learners,
-        wrapped into :class:`Orange.evaluation.reliability.Classifier`.
-    :type res: :class:`Orange.evaluation.testing.ExperimentResults`
-
-    Return Pearson's coefficient between the prediction error and each of the
-    used reliability estimates. Also, return the p-value of each of
-    the coefficients.
-    """
-    prediction_error = get_prediction_error_list(res)
-    results = []
-    for i in xrange(len(res.results[0].probabilities[0].reliability_estimate)):
-        reliability_estimate, signed_or_absolute, method = get_reliability_estimation_list(res, i)
-        try:
-            if signed_or_absolute == SIGNED:
-                r, p = statc.pearsonr(prediction_error, reliability_estimate)
-            else:
-                r, p = statc.pearsonr([abs(pe) for pe in prediction_error], reliability_estimate)
-        except Exception:
-            r = p = float("NaN")
-        results.append((r, p, signed_or_absolute, method))
-    return results
-
-def get_spearman_r(res):
-    """
-    :param res: results of evaluation, done using learners,
-        wrapped into :class:`Orange.evaluation.reliability.Classifier`.
-    :type res: :class:`Orange.evaluation.testing.ExperimentResults`
-
-    Return Spearman's coefficient between the prediction error and each of the
-    used reliability estimates. Also, return the p-value of each of
-    the coefficients.
-    """
-    prediction_error = get_prediction_error_list(res)
-    results = []
-    for i in xrange(len(res.results[0].probabilities[0].reliability_estimate)):
-        reliability_estimate, signed_or_absolute, method = get_reliability_estimation_list(res, i)
-        try:
-            if signed_or_absolute == SIGNED:
-                r, p = statc.spearmanr(prediction_error, reliability_estimate)
-            else:
-                r, p = statc.spearmanr([abs(pe) for pe in prediction_error], reliability_estimate)
-        except Exception:
-            r = p = float("NaN")
-        results.append((r, p, signed_or_absolute, method))
-    return results
-
-def get_pearson_r_by_iterations(res):
-    """
-    :param res: results of evaluation, done using learners,
-        wrapped into :class:`Orange.evaluation.reliability.Classifier`.
-    :type res: :class:`Orange.evaluation.testing.ExperimentResults`
-
-    Return average Pearson's coefficient over all folds between prediction error
-    and each of the used estimates.
-    """
-    results_by_fold = Orange.evaluation.scoring.split_by_iterations(res)
-    number_of_estimates = len(res.results[0].probabilities[0].reliability_estimate)
-    number_of_instances = len(res.results)
-    number_of_folds = len(results_by_fold)
-    results = [0 for _ in xrange(number_of_estimates)]
-    sig = [0 for _ in xrange(number_of_estimates)]
-    method_list = [0 for _ in xrange(number_of_estimates)]
-
-    for res in results_by_fold:
-        prediction_error = get_prediction_error_list(res)
-        for i in xrange(number_of_estimates):
-            reliability_estimate, signed_or_absolute, method = get_reliability_estimation_list(res, i)
-            try:
-                if signed_or_absolute == SIGNED:
-                    r, _ = statc.pearsonr(prediction_error, reliability_estimate)
-                else:
-                    r, _ = statc.pearsonr([abs(pe) for pe in prediction_error], reliability_estimate)
-            except Exception:
-                r = float("NaN")
-            results[i] += r
-            sig[i] = signed_or_absolute
-            method_list[i] = method
-
-    # Calculate p-values
-    results = [float(res) / number_of_folds for res in results]
-    ps = [p_value_from_r(r, number_of_instances) for r in results]
-
-    return zip(results, ps, sig, method_list)
-
-def p_value_from_r(r, n):
-    """
-    Calculate p-value from the paerson coefficient and the sample size.
-    """
-    df = n - 2
-    t = r * (df / ((-r + 1.0 + 1e-30) * (r + 1.0 + 1e-30))) ** 0.5
-    return statc.betai (df * 0.5, 0.5, df / (df + t * t))
-
-
-# Distances between two discrete probability distributions
-#TODO Document those.
-def normalize_both(p, q):
-    if not p.normalized:
-        p.normalize()
-    if not q.normalized:
-        q.normalize()
-    return p, q
-
-def minkowsky_dist(p, q, m=2):
-    p, q = normalize_both(p, q)
-    dist = 0
-    for i in range(len(p)):
-        dist += abs(p[i]-q[i])**m
-    return dist**(1./m)
-
-def manhattan_distance(p, q):
-    return minkowsky_dist(p, q, m=1)
-
-def euclidean_dist(p, q):
-    return minkowsky_dist(p, q, m=2)
-
-def variance_dist(p, q):
-    return euclidean_dist(p, q) ** 2
-
-def max_dist(p, q):
-    p, q = normalize_both(p, q)
-    return max([abs(p[i]-q[i]) for i in range(len(p))])
-
-def hellinger_dist(p, q):
-    p, q = normalize_both(p, q)
-    dist = 0
-    for i in range(len(p)):
-        dist += (math.sqrt(p[i])-math.sqrt(q[i])) ** 2
-    return dist
-
-def my_log(x):
-    return 0 if x == 0 else x * math.log(x)
-
-def kullback_leibler(p, q):
-    p, q = normalize_both(p, q)
-    dist = 0
-    for i in range(len(p)):
-        dist += my_log(p[i]-q[i])
-    return dist
-
-def cosine(p, q):
-    p, q = normalize_both(p, q)
-    p, q = [pp for pp in p], [qq for qq in q]
-    return 1 - numpy.dot(x,y) / (numpy.linalg.norm(p)*numpy.linalg.norm(q))
-
-
-class Estimate:
-    """
-    Reliability estimate. Contains attributes that describe the results of
-    reliability estimation.
-
-    .. attribute:: estimate
-
-        A numerical reliability estimate.
-
-    .. attribute:: signed_or_absolute
-
-        Determines whether the method used gives a signed or absolute result.
-        Has a value of either :obj:`SIGNED` or :obj:`ABSOLUTE`.
-
-    .. attribute:: method
-
-        An integer ID of reliability estimation method used.
-
-    .. attribute:: method_name
-
-        Name (string) of reliability estimation method used.
-
-    .. attribute:: icv_method
-
-        An integer ID of reliability estimation method that performed best,
-        as determined by ICV, and of which estimate is stored in the
-        :obj:`estimate` field. (:obj:`None` when ICV was not used.)
-
-    .. attribute:: icv_method_name
-
-        Name (string) of reliability estimation method that performed best,
-        as determined by ICV. (:obj:`None` when ICV was not used.)
-
-    """
-    def __init__(self, estimate, signed_or_absolute, method, icv_method= -1):
-        self.estimate = estimate
-        self.signed_or_absolute = signed_or_absolute
-        self.method = method
-        self.method_name = METHOD_NAME[method]
-        self.icv_method = icv_method
-        self.icv_method_name = METHOD_NAME[icv_method] if icv_method != -1 else ""
-        self.text_description = None
-
-class DescriptiveAnalysis:
-    def __init__(self, estimator, desc=["high", "medium", "low"], procentage=[0.00, 0.33, 0.66], name="da"):
-        self.desc = desc
-        self.procentage = procentage
-        self.estimator = estimator
-        self.name = name
-
-    def __call__(self, instances, weight=None, **kwds):
-
-        # Calculate borders using cross validation
-        res = Orange.evaluation.testing.cross_validation([self.estimator], instances)
-        all_borders = []
-        for i in xrange(len(res.results[0].probabilities[0].reliability_estimate)):
-            estimates, signed_or_absolute, method = get_reliability_estimation_list(res, i)
-            sorted_estimates = sorted(abs(x) for x in estimates)
-            borders = [sorted_estimates[int(len(estimates) * p) - 1]  for p in self.procentage]
-            all_borders.append(borders)
-
-        # Learn on whole train data
-        estimator_classifier = self.estimator(instances)
-
-        return DescriptiveAnalysisClassifier(estimator_classifier, all_borders, self.desc)
-
-class DescriptiveAnalysisClassifier:
-    def __init__(self, estimator_classifier, all_borders, desc):
-        self.estimator_classifier = estimator_classifier
-        self.all_borders = all_borders
-        self.desc = desc
-
-    def __call__(self, instance, result_type=Orange.core.GetValue):
-        predicted, probabilities = self.estimator_classifier(instance, Orange.core.GetBoth)
-
-        for borders, estimate in zip(self.all_borders, probabilities.reliability_estimate):
-            estimate.text_description = self.desc[0]
-            for lower_border, text_desc in zip(borders, self.desc):
-                if estimate.estimate >= lower_border:
-                    estimate.text_description = text_desc
-
-        # Return the appropriate type of result
-        if result_type == Orange.core.GetValue:
-            return predicted
-        elif result_type == Orange.core.GetProbabilities:
-            return probabilities
-        else:
-            return predicted, probabilities
-
-class SensitivityAnalysis:
-    """
-    
-    :param e: List of possible :math:`\epsilon` values for SAvar and SAbias
-        reliability estimates.
-    :type e: list of floats
-    
-    :rtype: :class:`Orange.evaluation.reliability.SensitivityAnalysisClassifier`
-    
-    To estimate the reliability of prediction for given instance,
-    the learning set is extended with this instance, labeled with
-    :math:`K + \epsilon (l_{max} - l_{min})`,
-    where :math:`K` denotes the initial prediction,
-    :math:`\epsilon` is sensitivity parameter and :math:`l_{min}` and
-    :math:`l_{max}` denote lower and the upper bound of the learning
-    instances' labels. After computing different sensitivity predictions
-    using different values of :math:`\epsilon`, the prediction are combined
-    into SAvar and SAbias. SAbias can be used in a signed or absolute form.
-
-    :math:`SAvar = \\frac{\sum_{\epsilon \in E}(K_{\epsilon} - K_{-\epsilon})}{|E|}`
-
-    :math:`SAbias = \\frac{\sum_{\epsilon \in E} (K_{\epsilon} - K ) + (K_{-\epsilon} - K)}{2 |E|}`
-    
-    
-    """
-    def __init__(self, e=[0.01, 0.1, 0.5, 1.0, 2.0], name="sa"):
-        self.e = e
-        self.name = name
-
-    def __call__(self, instances, learner):
-        min_value = max_value = instances[0].getclass().value
-        for ex in instances:
-            if ex.getclass().value > max_value:
-                max_value = ex.getclass().value
-            if ex.getclass().value < min_value:
-                min_value = ex.getclass().value
-        return SensitivityAnalysisClassifier(self.e, instances, min_value, max_value, learner)
-
-class SensitivityAnalysisClassifier:
-    def __init__(self, e, instances, min_value, max_value, learner):
-        self.e = e
-        self.instances = instances
-        self.max_value = max_value
-        self.min_value = min_value
-        self.learner = learner
-
-    def __call__(self, instance, predicted, probabilities):
-        # Create new dataset
-        r_data = Orange.data.Table(self.instances)
-
-        # Create new instance
-        modified_instance = Orange.data.Instance(instance)
-
-        # Append it to the data
-        r_data.append(modified_instance)
-
-        # Calculate SAvar & SAbias
-        SAvar = SAbias = 0
-
-        for eps in self.e:
-            # +epsilon
-            r_data[-1].setclass(predicted.value + eps * (self.max_value - self.min_value))
-            c = self.learner(r_data)
-            k_plus = c(instance, Orange.core.GetValue)
-
-            # -epsilon
-            r_data[-1].setclass(predicted.value - eps * (self.max_value - self.min_value))
-            c = self.learner(r_data)
-            k_minus = c(instance, Orange.core.GetValue)
-            #print len(r_data)
-            #print eps*(self.max_value - self.min_value)
-            #print k_plus
-            #print k_minus
-            # calculate part SAvar and SAbias
-            SAvar += k_plus.value - k_minus.value
-            SAbias += k_plus.value + k_minus.value - 2 * predicted.value
-
-        SAvar /= len(self.e)
-        SAbias /= 2 * len(self.e)
-
-        return [Estimate(SAvar, ABSOLUTE, SAVAR_ABSOLUTE),
-                Estimate(SAbias, SIGNED, SABIAS_SIGNED),
-                Estimate(abs(SAbias), ABSOLUTE, SABIAS_ABSOLUTE)]
-
-
-
-class ReferenceExpectedError:
-    """
-
-    :rtype: :class:`Orange.evaluation.reliability.ReferenceExpectedErrorClassifier`
-
-    Reference reliability estimation method for classification as used in Evaluating Reliability of Single
-    Classifications of Neural Networks, Darko Pevec, 2011.
-
-    :math:`O_{ref} = 2 (\hat y - \hat y ^2) = 2 \hat y (1-\hat y)`
-
-    where :math:`\hat y` is the estimated probability of the predicted class.
-
-    Note that for this method, in contrast with all others, a greater estimate means lower reliability (greater
-    expected error).
-
-    """
-    def __init__(self, name="reference"):
-        self.name = name
-
-    def __call__(self, instances, learner):
-        classifier = learner(instances)
-        return ReferenceExpectedErrorClassifier(classifier)
-
-    
-class ReferenceExpectedErrorClassifier:
-
-    def __init__(self, classifier):
-        self.classifier = classifier
-
-    def __call__(self, instance, *args):
-        y_hat = max(self.classifier(instance, Orange.classification.Classifier.GetProbabilities))
-        return [Estimate(2 * y_hat * (1 - y_hat), ABSOLUTE, ERR_ABSOLUTE)]
-
-    
-
-class BaggingVariance:
-    """
-    
-    :param m: Number of bagging models to be used with BAGV estimate
-    :type m: int
-    
-    :rtype: :class:`Orange.evaluation.reliability.BaggingVarianceClassifier`
-    
-    :math:`m` different bagging models are constructed and used to estimate
-    the value of dependent variable for a given instance. In regression,
-    the variance of those predictions is used as a prediction reliability
-    estimate.
-
-    :math:`BAGV = \\frac{1}{m} \sum_{i=1}^{m} (K_i - K)^2`
-
-    where :math:`K = \\frac{\sum_{i=1}^{m} K_i}{m}` and :math:`K_i` are
-    predictions of individual constructed models. Note that a greater value
-    implies greater error.
-
-    For classification, 1 minus the average Euclidean distance between class
-    probability distributions predicted by the model, and distributions
-    predicted by the individual bagged models, is used as the BAGV reliability
-    measure. Note that in this case a greater value implies a better
-    prediction.
-    
-    """
-    def __init__(self, m=50, name="bv"):
-        self.m = m
-        self.name = name
-
-    def __call__(self, instances, learner):
-        classifiers = []
-
-        if instances.domain.class_var.var_type == Orange.feature.Descriptor.Discrete:
-            classifier = learner(instances)
-        else:
-            classifier = None
-
-        # Create bagged classifiers using sampling with replacement
-        for _ in xrange(self.m):
-            selection = select_with_repeat(len(instances))
-            data = instances.select(selection)
-            classifiers.append(learner(data))
-        return BaggingVarianceClassifier(classifiers, classifier)
-
-class BaggingVarianceClassifier:
-    def __init__(self, classifiers, classifier=None):
-        self.classifiers = classifiers
-        self.classifier = classifier
-
-    def __call__(self, instance, *args):
-        BAGV = 0
-
-        # Calculate the bagging variance
-        if instance.domain.class_var.var_type == Orange.feature.Descriptor.Continuous:
-            bagged_values = [c(instance, Orange.core.GetValue).value for c in self.classifiers if c is not None]
-        elif instance.domain.class_var.var_type == Orange.feature.Descriptor.Discrete:
-            estimate = self.classifier(instance, Orange.core.GetProbabilities)
-            bagged_values = [euclidean_dist(c(instance, Orange.core.GetProbabilities), estimate) for c in self.classifiers if c is not None]
-        k = sum(bagged_values) / len(bagged_values)
-
-        BAGV = sum((bagged_value - k) ** 2 for bagged_value in bagged_values) / len(bagged_values)
-        if instance.domain.class_var.var_type == Orange.feature.Descriptor.Discrete:
-            BAGV = 1 - BAGV
-
-        return [Estimate(BAGV, ABSOLUTE, BAGV_ABSOLUTE)]
-
-class LocalCrossValidation:
-    """
-
-    :param k: Number of nearest neighbours used in LCV estimate
-    :type k: int
-
-    :param distance: function that computes a distance between two discrete
-        distributions (used only in classification problems). The default
-        is Hellinger distance.
-    :type distance: function
-
-    :param distance_weighted: for classification reliability estimation,
-        use an average distance between distributions, weighted by :math:`e^{-d}`,
-        where :math:`d` is the distance between predicted instance and the
-        neighbour.
-
-    :rtype: :class:`Orange.evaluation.reliability.LocalCrossValidationClassifier`
-
-    :math:`k` nearest neighbours to the given instance are found and put in
-    a separate data set. On this data set, a leave-one-out validation is
-    performed. Reliability estimate for regression is then the distance
-    weighted absolute prediction error. In classification, 1 minus the average
-    distance between the predicted class probability distribution and the
-    (trivial) probability distributions of the nearest neighbour.
-
-    If a special value 0 is passed as :math:`k` (as is by default),
-    it is set as 1/20 of data set size (or 5, whichever is greater).
-
-    Summary of the algorithm for regression:
-
-    1. Determine the set of k nearest neighours :math:`N = { (x_1, c_1),...,
-       (x_k, c_k)}`.
-    2. On this set, compute leave-one-out predictions :math:`K_i` and
-       prediction errors :math:`E_i = | C_i - K_i |`.
-    3. :math:`LCV(x) = \\frac{ \sum_{(x_i, c_i) \in N} d(x_i, x) * E_i }{ \sum_{(x_i, c_i) \in N} d(x_i, x) }`
-
-    """
-    def __init__(self, k=0, distance=hellinger_dist, distance_weighted=True, name="lcv"):
-        self.k = k
-        self.distance = distance
-        self.distance_weighted = distance_weighted
-        self.name = name
-
-    def __call__(self, instances, learner):
-        nearest_neighbours_constructor = Orange.classification.knn.FindNearestConstructor()
-        nearest_neighbours_constructor.distanceConstructor = Orange.distance.Euclidean()
-
-        distance_id = Orange.feature.Descriptor.new_meta_id()
-        nearest_neighbours = nearest_neighbours_constructor(instances, 0, distance_id)
-
-        if self.k == 0:
-            self.k = max(5, len(instances) / 20)
-
-        return LocalCrossValidationClassifier(distance_id, nearest_neighbours, self.k, learner,
-            distance=self.distance, distance_weighted=self.distance_weighted)
-
-class LocalCrossValidationClassifier:
-    def __init__(self, distance_id, nearest_neighbours, k, learner, **kwds):
-        self.distance_id = distance_id
-        self.nearest_neighbours = nearest_neighbours
-        self.k = k
-        self.learner = learner
-        for a,b in kwds.items():
-            setattr(self, a, b)
-
-    def __call__(self, instance, *args):
-        LCVer = 0
-        LCVdi = 0
-
-        # Find k nearest neighbors
-
-        knn = [ex for ex in self.nearest_neighbours(instance, self.k)]
-
-        # leave one out of prediction error
-        for i in xrange(len(knn)):
-            train = knn[:]
-            del train[i]
-
-            classifier = self.learner(Orange.data.Table(train))
-
-            if instance.domain.class_var.var_type == Orange.feature.Descriptor.Continuous:
-                returned_value = classifier(knn[i], Orange.core.GetValue)
-                e = abs(knn[i].getclass().value - returned_value.value)
-
-            elif instance.domain.class_var.var_type == Orange.feature.Descriptor.Discrete:
-                returned_value = classifier(knn[i], Orange.core.GetProbabilities)
-                probabilities = [knn[i].get_class() == val for val in instance.domain.class_var.values]
-                e = self.distance(returned_value, Orange.statistics.distribution.Discrete(probabilities))
-
-            dist = math.exp(-knn[i][self.distance_id]) if self.distance_weighted else 1.0
-            LCVer += e * dist
-            LCVdi += dist
-
-        LCV = LCVer / LCVdi if LCVdi != 0 else 0
-        if math.isnan(LCV):
-            LCV = 0.0
-
-        if instance.domain.class_var.var_type == Orange.feature.Descriptor.Discrete:
-            LCV = 1 - LCV
-
-        return [ Estimate(LCV, ABSOLUTE, LCV_ABSOLUTE) ]
-
-class CNeighbours:
-    """
-    
-    :param k: Number of nearest neighbours used in CNK estimate
-    :type k: int
-
-    :param distance: function that computes a distance between two discrete
-        distributions (used only in classification problems). The default
-        is Hellinger distance.
-    :type distance: function
-    
-    :rtype: :class:`Orange.evaluation.reliability.CNeighboursClassifier`
-    
-    For regression, CNK is defined for an unlabeled instance as a difference
-    between average label of its nearest neighbours and its prediction. CNK
-    can be used as a signed or absolute estimate.
-    
-    :math:`CNK = \\frac{\sum_{i=1}^{k}C_i}{k} - K`
-    
-    where :math:`k` denotes number of neighbors, C :sub:`i` denotes neighbours'
-    labels and :math:`K` denotes the instance's prediction. Note that a greater
-    value implies greater prediction error.
-
-    For classification, CNK is equal to 1 minus the average distance between
-    predicted class distribution and (trivial) class distributions of the
-    $k$ nearest neighbours from the learning set. Note that in this case
-    a greater value implies better prediction.
-    
-    """
-    def __init__(self, k=5, distance=hellinger_dist, name = "cnk"):
-        self.k = k
-        self.distance = distance
-        self.name = name
-
-    def __call__(self, instances, learner):
-        nearest_neighbours_constructor = Orange.classification.knn.FindNearestConstructor()
-        nearest_neighbours_constructor.distanceConstructor = Orange.distance.Euclidean()
-
-        distance_id = Orange.feature.Descriptor.new_meta_id()
-        nearest_neighbours = nearest_neighbours_constructor(instances, 0, distance_id)
-        return CNeighboursClassifier(nearest_neighbours, self.k, distance=self.distance)
-
-class CNeighboursClassifier:
-    def __init__(self, nearest_neighbours, k, distance):
-        self.nearest_neighbours = nearest_neighbours
-        self.k = k
-        self.distance = distance
-
-    def __call__(self, instance, predicted, probabilities):
-        CNK = 0
-
-        # Find k nearest neighbors
-
-        knn = [ex for ex in self.nearest_neighbours(instance, self.k)]
-
-        # average label of neighbors
-        if ex.domain.class_var.var_type == Orange.feature.Descriptor.Continuous:
-            for ex in knn:
-                CNK += ex.getclass().value
-            CNK /= self.k
-            CNK -= predicted.value
-
-            return [Estimate(CNK, SIGNED, CNK_SIGNED),
-                    Estimate(abs(CNK), ABSOLUTE, CNK_ABSOLUTE)]
-        elif ex.domain.class_var.var_type == Orange.feature.Descriptor.Discrete:
-            knn_l = Orange.classification.knn.kNNLearner(k=self.k)
-            knn_c = knn_l(knn)
-            for ex in knn:
-                CNK -= self.distance(probabilities, knn_c(ex, Orange.classification.Classifier.GetProbabilities))
-            CNK /= self.k
-            CNK += 1
-
-            return [Estimate(CNK, ABSOLUTE, CNK_ABSOLUTE)]
-
-class Mahalanobis:
-    """
-    
-    :param k: Number of nearest neighbours used in Mahalanobis estimate.
-    :type k: int
-    
-    :rtype: :class:`Orange.evaluation.reliability.MahalanobisClassifier`
-    
-    Mahalanobis distance reliability estimate is defined as
-    `mahalanobis distance <http://en.wikipedia.org/wiki/Mahalanobis_distance>`_
-    to the evaluated instance's :math:`k` nearest neighbours.
-
-    
-    """
-    def __init__(self, k=3, name="mahalanobis"):
-        self.k = k
-        self.name = name
-
-    def __call__(self, instances, *args):
-        nnm = Orange.classification.knn.FindNearestConstructor()
-        nnm.distanceConstructor = Orange.distance.Mahalanobis()
-
-        mid = Orange.feature.Descriptor.new_meta_id()
-        nnm = nnm(instances, 0, mid)
-        return MahalanobisClassifier(self.k, nnm, mid)
-
-class MahalanobisClassifier:
-    def __init__(self, k, nnm, mid):
-        self.k = k
-        self.nnm = nnm
-        self.mid = mid
-
-    def __call__(self, instance, *args):
-        mahalanobis_distance = 0
-
-        mahalanobis_distance = sum(ex[self.mid].value for ex in self.nnm(instance, self.k))
-
-        return [ Estimate(mahalanobis_distance, ABSOLUTE, MAHAL_ABSOLUTE) ]
-
-class MahalanobisToCenter:
-    """
-    :rtype: :class:`Orange.evaluation.reliability.MahalanobisToCenterClassifier`
-    
-    Mahalanobis distance to center reliability estimate is defined as a
-    `mahalanobis distance <http://en.wikipedia.org/wiki/Mahalanobis_distance>`_
-    between the predicted instance and the centroid of the data.
-
-    
-    """
-    def __init__(self, name="mahalanobis to center"):
-        self.name = name
-
-    def __call__(self, instances, *args):
-        dc = Orange.core.DomainContinuizer()
-        dc.classTreatment = Orange.core.DomainContinuizer.Ignore
-        dc.continuousTreatment = Orange.core.DomainContinuizer.NormalizeBySpan
-        dc.multinomialTreatment = Orange.core.DomainContinuizer.NValues
-
-        new_domain = dc(instances)
-        new_instances = instances.translate(new_domain)
-
-        X, _, _ = new_instances.to_numpy()
-        instance_avg = numpy.average(X, 0)
-
-        distance_constructor = Orange.distance.Mahalanobis()
-        distance = distance_constructor(new_instances)
-
-        average_instance = Orange.data.Instance(new_instances.domain, list(instance_avg) + ["?"])
-
-        return MahalanobisToCenterClassifier(distance, average_instance, new_domain)
-
-class MahalanobisToCenterClassifier:
-    def __init__(self, distance, average_instance, new_domain):
-        self.distance = distance
-        self.average_instance = average_instance
-        self.new_domain = new_domain
-
-    def __call__(self, instance, *args):
-
-        inst = Orange.data.Instance(self.new_domain, instance)
-
-        mahalanobis_to_center = self.distance(inst, self.average_instance)
-
-        return [ Estimate(mahalanobis_to_center, ABSOLUTE, MAHAL_TO_CENTER_ABSOLUTE) ]
-
-
-class BaggingVarianceCNeighbours:
-    """
-    
-    :param bagv: Instance of Bagging Variance estimator.
-    :type bagv: :class:`BaggingVariance`
-    
-    :param cnk: Instance of CNK estimator.
-    :type cnk: :class:`CNeighbours`
-    
-    :rtype: :class:`Orange.evaluation.reliability.BaggingVarianceCNeighboursClassifier`
-    
-    BVCK is a combination (average) of Bagging variance and local modeling of
-    prediction error.
-    
-    """
-    def __init__(self, bagv=BaggingVariance(), cnk=CNeighbours(), name="bvck"):
-        self.bagv = bagv
-        self.cnk = cnk
-        self.name = "bvck"
-
-    def __call__(self, instances, learner):
-        bagv_classifier = self.bagv(instances, learner)
-        cnk_classifier = self.cnk(instances, learner)
-        return BaggingVarianceCNeighboursClassifier(bagv_classifier, cnk_classifier)
-
-class BaggingVarianceCNeighboursClassifier:
-    def __init__(self, bagv_classifier, cnk_classifier):
-        self.bagv_classifier = bagv_classifier
-        self.cnk_classifier = cnk_classifier
-
-    def __call__(self, instance, predicted, probabilities):
-        bagv_estimates = self.bagv_classifier(instance, predicted, probabilities)
-        cnk_estimates = self.cnk_classifier(instance, predicted, probabilities)
-
-        bvck_value = (bagv_estimates[0].estimate + cnk_estimates[1].estimate) / 2
-        bvck_estimates = [ Estimate(bvck_value, ABSOLUTE, BVCK_ABSOLUTE) ]
-        bvck_estimates.extend(bagv_estimates)
-        bvck_estimates.extend(cnk_estimates)
-        return bvck_estimates
-
-class ErrorPredicting:
-    def __init__(self, name = "ep"):
-        self.name = name
-
-    def __call__(self, instances, learner):
-        res = Orange.evaluation.testing.cross_validation([learner], instances)
-        prediction_errors = get_prediction_error_list(res)
-
-        new_domain = Orange.data.Domain(instances.domain.attributes, Orange.core.FloatVariable("pe"))
-        new_dataset = Orange.data.Table(new_domain, instances)
-
-        for instance, prediction_error in izip(new_dataset, prediction_errors):
-            instance.set_class(prediction_error)
-
-        rf = Orange.ensemble.forest.RandomForestLearner()
-        rf_classifier = rf(new_dataset)
-
-        return ErrorPredictingClassification(rf_classifier, new_domain)
-
-class ErrorPredictingClassification:
-    def __init__(self, rf_classifier, new_domain):
-        self.rf_classifier = rf_classifier
-        self.new_domain = new_domain
-
-    def __call__(self, instance, predicted, probabilities):
-        new_instance = Orange.data.Instance(self.new_domain, instance)
-        value = self.rf_classifier(new_instance, Orange.core.GetValue)
-
-        return [Estimate(value.value, SIGNED, SABIAS_SIGNED)]
-
-def gauss_kernel(x, sigma=1):
-    return 1./(sigma*math.sqrt(2*math.pi)) * math.exp(-1./2*(x/sigma)**2)
-
-class ParzenWindowDensityBased:
-    """
-    :param K: kernel function. Default: gaussian.
-    :type K: function
-
-    :param d_measure: distance measure for inter-instance distance.
-    :type d_measure: :class:`Orange.distance.DistanceConstructor`
-
-    :rtype: :class:`Orange.evaluation.reliability.ParzenWindowDensityBasedClassifier`
-
-    Returns a value that estimates a density of problem space around the
-    instance being predicted.
-    """
-    def __init__(self, K=gauss_kernel, d_measure=Orange.distance.Euclidean(), name="density"):
-        self.K = K
-        self.d_measure = d_measure
-        self.name = name
-
-    def __call__(self, instances, learner):
-
-        self.distance = self.d_measure(instances)
-
-        def density(x):
-            l, dens = len(instances), 0
-            for ex in instances:
-                dens += self.K(self.distance(x,ex))
-            return dens / l
-
-        max_density = max([density(ex) for ex in instances])
-
-        return ParzenWindowDensityBasedClassifier(density, max_density)
-
-class ParzenWindowDensityBasedClassifier:
-
-    def __init__(self, density, max_density):
-        self.density = density
-        self.max_density = max_density
-
-
-    def __call__(self, instance, *args):
-
-        DENS = self.max_density-self.density(instance)
-
-        return [Estimate(DENS, ABSOLUTE, DENS_ABSOLUTE)]
-
-class Learner:
-    """
-    Reliability estimation wrapper around a learner we want to test.
-    Different reliability estimation algorithms can be used on the
-    chosen learner. This learner works as any other and can be used as one,
-    but it returns the classifier, wrapped into an instance of
-    :class:`Orange.evaluation.reliability.Classifier`.
-    
-    :param box_learner: Learner we want to wrap into a reliability estimation
-        classifier.
-    :type box_learner: :obj:`~Orange.classification.Learner`
-    
-    :param estimators: List of different reliability estimation methods we
-                       want to use on the chosen learner.
-    :type estimators: :obj:`list` of reliability estimators
-    
-    :param name: Name of this reliability learner
-    :type name: string
-    
-    :rtype: :class:`Orange.evaluation.reliability.Learner`
-    """
-    def __init__(self, box_learner, name="Reliability estimation",
-                 estimators=[SensitivityAnalysis(),
-                             LocalCrossValidation(),
-                             BaggingVarianceCNeighbours(),
-                             Mahalanobis(),
-                             MahalanobisToCenter()],
-                 **kwds):
-        self.__dict__.update(kwds)
-        self.name = name
-        self.estimators = estimators
-        self.box_learner = box_learner
-        self.blending = False
-
-
-    def __call__(self, instances, weight=None, **kwds):
-        """Learn from the given table of data instances.
-        
-        :param instances: Data instances to learn from.
-        :type instances: Orange.data.Table
-        :param weight: Id of meta attribute with weights of instances
-        :type weight: int
-        :rtype: :class:`Orange.evaluation.reliability.Classifier`
-        """
-
-        blending_classifier = None
-        new_domain = None
-
-#        if instances.domain.class_var.var_type != Orange.feature.Continuous.Continuous:
-#            raise Exception("This method only works on data with continuous class.")
-
-        return Classifier(instances, self.box_learner, self.estimators, self.blending, new_domain, blending_classifier)
-
-    def internal_cross_validation(self, instances, folds=10):
-        """ Perform the internal cross validation for getting the best
-        reliability estimate. It uses the reliability estimators defined in
-        estimators attribute.
-
-        Returns the id of the method that scored the best.
-
-        :param instances: Data instances to use for ICV.
-        :type instances: :class:`Orange.data.Table`
-        :param folds: number of folds for ICV.
-        :type folds: int
-        :rtype: int
-
-        """
-        res = Orange.evaluation.testing.cross_validation([self], instances, folds=folds)
-        results = get_pearson_r(res)
-        sorted_results = sorted(results)
-        return sorted_results[-1][3]
-
-    def internal_cross_validation_testing(self, instances, folds=10):
-        """ Perform internal cross validation (as in Automatic selection of
-        reliability estimates for individual regression predictions,
-        Zoran Bosnic, 2010) and return id of the method
-        that scored best on this data.
-
-        :param instances: Data instances to use for ICV.
-        :type instances: :class:`Orange.data.Table`
-        :param folds: number of folds for ICV.
-        :type folds: int
-        :rtype: int
-
-        """
-        cv_indices = Orange.core.MakeRandomIndicesCV(instances, folds)
-
-        list_of_rs = []
-
-        sum_of_rs = defaultdict(float)
-
-        for fold in xrange(folds):
-            data = instances.select(cv_indices, fold)
-            if len(data) < 10:
-                res = Orange.evaluation.testing.leave_one_out([self], data)
-            else:
-                res = Orange.evaluation.testing.cross_validation([self], data)
-            results = get_pearson_r(res)
-            for r, _, _, method in results:
-                sum_of_rs[method] += r
-        sorted_sum_of_rs = sorted(sum_of_rs.items(), key=lambda estimate: estimate[1], reverse=True)
-        return sorted_sum_of_rs[0][0]
-
-    labels = ["SAvar", "SAbias", "BAGV", "CNK", "LCV", "BVCK", "Mahalanobis", "ICV"]
-
-class Classifier:
-    """
-    A reliability estimation wrapper for classifiers.
-
-    What distinguishes this classifier is that the returned probabilities (if
-    :obj:`Orange.classification.Classifier.GetProbabilities` or
-    :obj:`Orange.classification.Classifier.GetBoth` is passed) contain an
-    additional attribute :obj:`reliability_estimate`, which is an instance of
-    :class:`~Orange.evaluation.reliability.Estimate`.
-
-    """
-
-    def __init__(self, instances, box_learner, estimators, blending, blending_domain, rf_classifier, **kwds):
-        self.__dict__.update(kwds)
-        self.instances = instances
-        self.box_learner = box_learner
-        self.estimators = estimators
-        self.blending = blending
-        self.blending_domain = blending_domain
-        self.rf_classifier = rf_classifier
-
-        # Train the learner with original data
-        self.classifier = box_learner(instances)
-
-        # Train all the estimators and create their classifiers
-        self.estimation_classifiers = [estimator(instances, box_learner) for estimator in estimators]
-
-    def __call__(self, instance, result_type=Orange.core.GetValue):
-        """
-        Classify and estimate reliability of estimation for a new instance.
-        When :obj:`result_type` is set to
-        :obj:`Orange.classification.Classifier.GetBoth` or
-        :obj:`Orange.classification.Classifier.GetProbabilities`,
-        an additional attribute :obj:`reliability_estimate`,
-        which is an instance of
-        :class:`~Orange.evaluation.reliability.Estimate`,
-        is added to the distribution object.
-        
-        :param instance: instance to be classified.
-        :type instance: :class:`Orange.data.Instance`
-        :param result_type: :class:`Orange.classification.Classifier.GetValue` or \
-              :class:`Orange.classification.Classifier.GetProbabilities` or
-              :class:`Orange.classification.Classifier.GetBoth`
-        
-        :rtype: :class:`Orange.data.Value`, 
-              :class:`Orange.statistics.Distribution` or a tuple with both
-        """
-        predicted, probabilities = self.classifier(instance, Orange.core.GetBoth)
-
-        # Create a place holder for estimates
-        if probabilities is None:
-            probabilities = Orange.statistics.distribution.Continuous()
-        #with warnings.catch_warnings():
-        #    warnings.simplefilter("ignore")
-        probabilities.setattr('reliability_estimate', [])
-
-        # Calculate all the estimates and add them to the results
-        for estimate in self.estimation_classifiers:
-            probabilities.reliability_estimate.extend(estimate(instance, predicted, probabilities))
-
-        # Return the appropriate type of result
-        if result_type == Orange.core.GetValue:
-            return predicted
-        elif result_type == Orange.core.GetProbabilities:
-            return probabilities
-        else:
-            return predicted, probabilities
-
-# Functions for testing and plotting
-#TODO Document those.
-def get_acc_rel(method, data, learner):
-    estimators = [method]
-    reliability = Orange.evaluation.reliability.Learner(learner, estimators=estimators)
-    #results = Orange.evaluation.testing.leave_one_out([reliability], data)
-    results = Orange.evaluation.testing.cross_validation([reliability], data)
-
-    rels, acc = [], []
-
-    for res in results.results:
-        rels.append(res.probabilities[0].reliability_estimate[0].estimate)
-        acc.append(res.probabilities[0][res.actual_class])
-
-    return rels, acc
-
-
-def rel_acc_plot(rels, acc, file_name=None, colors=None):
-
-    import matplotlib.pylab as plt
-    
-    if colors is None:
-        colors = "k"
-    plt.scatter(rels, acc, c=colors)
-    plt.xlim(0.,1.)
-    plt.ylim(ymin=0.)
-    plt.xlabel("Reliability")
-    plt.ylabel("Accuracy")
-    if file_name is None:
-        plt.show()
-    else:
-        plt.savefig(file_name)
-
-def rel_acc_compute_plot(method, data, learner, file_name=None, colors=None):
-
-    plt.clf()
-
-    rels, acc = get_acc_rel(method, data, learner)
-    el_acc_plot(acc, rels, file_name=file_name, colors=colors)
-    
-
-def acc_rel_correlation(method, data, learner):
-    import scipy.stats
-    rels, acc = get_acc_rel(method, data, learner)
-    return scipy.stats.spearmanr(acc, rels)[0]

File _reliability/widgets/OWReliability.py

-"""
-<name>Reliability</name>
-<contact>Ales Erjavec (ales.erjavec(@at@)fri.uni-lj.si)</contact>
-<priority>310</priority>
-<icon>icons/Reliability.svg</icon>
-"""
-
-from __future__ import absolute_import
-
-import Orange
-import _reliability as reliability
-from Orange.evaluation import testing
-#from Orange.utils import progress_bar_milestones
-from functools import partial
- 
-from OWWidget import *
-import OWGUI
-
-class OWReliability(OWWidget):
-    settingsList = ["variance_checked", "bias_checked", "bagged_variance",
-        "local_cv", "local_model_pred_error", "bagging_variance_cn", 
-        "mahalanobis_distance", "var_e", "bias_e", "bagged_m", "local_cv_k",
-        "local_pe_k", "bagged_cn_m", "bagged_cn_k", "mahalanobis_k",
-        "include_error", "include_class", "include_input_features",
-        "auto_commit"]
-    
-    def __init__(self, parent=None, signalManager=None, title="Reliability"):
-        OWWidget.__init__(self, parent, signalManager, title, wantMainArea=False)
-        
-        self.inputs = [("Learner", Orange.core.Learner, self.set_learner),
-                       ("Training Data", Orange.data.Table, self.set_train_data),
-                       ("Test Data", Orange.data.Table, self.set_test_data)]
-        
-        self.outputs = [("Reliability Scores", Orange.data.Table)]
-        
-        self.variance_checked = False
-        self.bias_checked = False
-        self.bagged_variance = False
-        self.local_cv = False
-        self.local_model_pred_error = False
-        self.bagging_variance_cn = False
-        self.mahalanobis_distance = True
-        
-        self.var_e = "0.01, 0.1, 0.5, 1.0, 2.0"
-        self.bias_e =  "0.01, 0.1, 0.5, 1.0, 2.0"
-        self.bagged_m = 10
-        self.local_cv_k = 2
-        self.local_pe_k = 5
-        self.bagged_cn_m = 5
-        self.bagged_cn_k = 1
-        self.mahalanobis_k = 3
-        
-        self.include_error = True
-        self.include_class = True
-        self.include_input_features = False
-        self.auto_commit = False
-        
-        # (selected attr name, getter function, count of returned estimators, indices of estimator results to use)
-        self.estimators = \
-            [("variance_checked", self.get_SAVar, 3, [0]),
-             ("bias_checked", self.get_SABias, 3, [1, 2]),
-             ("bagged_variance", self.get_BAGV, 1, [0]),
-             ("local_cv", self.get_LCV, 1, [0]),
-             ("local_model_pred_error", self.get_CNK, 2, [0, 1]),
-             ("bagging_variance_cn", self.get_BVCK, 4, [0]),
-             ("mahalanobis_distance", self.get_Mahalanobis, 1, [0])]
-        
-        #####
-        # GUI
-        #####
-        self.loadSettings()
-        
-        box = OWGUI.widgetBox(self.controlArea, "Info", addSpace=True)
-        self.info_box = OWGUI.widgetLabel(box, "\n\n")
-        
-        rbox = OWGUI.widgetBox(self.controlArea, "Methods", addSpace=True)
-        def method_box(parent, name, value):
-            box = OWGUI.widgetBox(rbox, name, flat=False)
-            box.setCheckable(True)
-            box.setChecked(bool(getattr(self, value)))
-            self.connect(box, SIGNAL("toggled(bool)"),
-                         lambda on: (setattr(self, value, on),
-                                     self.method_selection_changed(value)))
-            return box
-            
-        e_validator = QRegExpValidator(QRegExp(r"\s*(-?[0-9]+(\.[0-9]*)\s*,\s*)+"), self)
-        variance_box = method_box(rbox, "Sensitivity analysis (variance)",
-                                  "variance_checked")
-        OWGUI.lineEdit(variance_box, self, "var_e", "Sensitivities:", 
-                       tooltip="List of possible e values (comma separated) for SAvar reliability estimates.", 
-                       callback=partial(self.method_param_changed, 0),
-                       validator=e_validator)
-        
-        bias_box = method_box(rbox, "Sensitivity analysis (bias)",
-                                    "bias_checked")
-        OWGUI.lineEdit(bias_box, self, "bias_e", "Sensitivities:", 
-                       tooltip="List of possible e values (comma separated) for SAbias reliability estimates.", 
-                       callback=partial(self.method_param_changed, 1),
-                       validator=e_validator)
-        
-        bagged_box = method_box(rbox, "Variance of bagged models",
-                                "bagged_variance")
-        
-        OWGUI.spin(bagged_box, self, "bagged_m", 2, 100, step=1,
-                   label="Models:",
-                   tooltip="Number of bagged models to be used with BAGV estimate.",
-                   callback=partial(self.method_param_changed, 2),
-                   keyboardTracking=False)
-        
-        local_cv_box = method_box(rbox, "Local cross validation",
-                                  "local_cv")
-        
-        OWGUI.spin(local_cv_box, self, "local_cv_k", 2, 20, step=1,
-                   label="Nearest neighbors:",
-                   tooltip="Number of nearest neighbors used in LCV estimate.",
-                   callback=partial(self.method_param_changed, 3),
-                   keyboardTracking=False)
-        
-        local_pe = method_box(rbox, "Local modeling of prediction error",
-                              "local_model_pred_error")
-        
-        OWGUI.spin(local_pe, self, "local_pe_k", 1, 20, step=1,
-                   label="Nearest neighbors:",
-                   tooltip="Number of nearest neighbors used in CNK estimate.",
-                   callback=partial(self.method_param_changed, 4),
-                   keyboardTracking=False)
-        
-        bagging_cnn = method_box(rbox, "Bagging variance c-neighbors",
-                                 "bagging_variance_cn")
-        
-        OWGUI.spin(bagging_cnn, self, "bagged_cn_m", 2, 100, step=1,
-                   label="Models:",
-                   tooltip="Number of bagged models to be used with BVCK estimate.",
-                   callback=partial(self.method_param_changed, 5),
-                   keyboardTracking=False)
-        
-        OWGUI.spin(bagging_cnn, self, "bagged_cn_k", 1, 20, step=1,
-                   label="Nearest neighbors:",
-                   tooltip="Number of nearest neighbors used in BVCK estimate.",
-                   callback=partial(self.method_param_changed, 5),
-                   keyboardTracking=False)
-        
-        mahalanobis_box = method_box(rbox, "Mahalanobis distance",
-                                     "mahalanobis_distance")
-        OWGUI.spin(mahalanobis_box, self, "mahalanobis_k", 1, 20, step=1,
-                   label="Nearest neighbors:",
-                   tooltip="Number of nearest neighbors used in BVCK estimate.",
-                   callback=partial(self.method_param_changed, 6),
-                   keyboardTracking=False)
-        
-        box = OWGUI.widgetBox(self.controlArea, "Output")
-        
-        OWGUI.checkBox(box, self, "include_error", "Include prediction error",
-                       tooltip="Include prediction error in the output",
-                       callback=self.commit_if)
-        
-        OWGUI.checkBox(box, self, "include_class", "Include original class and prediction",
-                       tooltip="Include original class and prediction in the output.",
-                       callback=self.commit_if)
-        
-        OWGUI.checkBox(box, self, "include_input_features", "Include input features",
-                       tooltip="Include features from the input data set.",
-                       callback=self.commit_if)
-        
-        cb = OWGUI.checkBox(box, self, "auto_commit", "Commit on any change",
-                            callback=self.commit_if)
-        
-        self.commit_button = b = OWGUI.button(box, self, "Commit",
-                                              callback=self.commit,
-                                              autoDefault=True)
-        
-        OWGUI.setStopper(self, b, cb, "output_changed", callback=self.commit)
-        
-        self.commit_button.setEnabled(any([getattr(self, selected) \
-                                for selected, _, _, _ in  self.estimators]))
-        
-        self.learner = None
-        self.train_data = None
-        self.test_data = None
-        self.output_changed = False
-        self.train_data_has_no_class = False
-        self.train_data_has_discrete_class = False
-        self.invalidate_results()
-        
-    def set_train_data(self, data=None):
-        self.error()
-        self.train_data_has_no_class = False
-        self.train_data_has_discrete_class = False
-        
-        if data is not None:
-            if not self.isDataWithClass(data, Orange.core.VarTypes.Continuous):
-                if not data.domain.class_var:
-                    self.train_data_has_no_class = True
-                elif not isinstance(data.domain.class_var,
-                                    Orange.feature.Continuous):
-                    self.train_data_has_discrete_class = True
-                    
-                data = None
-        
-        self.train_data = data
-        self.invalidate_results() 
-        
-    def set_test_data(self, data=None):
-        self.test_data = data
-        self.invalidate_results()
-        
-    def set_learner(self, learner=None):
-        self.learner = learner
-        self.invalidate_results()
-        
-    def handleNewSignals(self):
-        name = "No learner on input"
-        train = "No train data on input"
-        test = "No test data on input"
-        
-        if self.learner:
-            name = "Learner: " + (getattr(self.learner, "name") or type(self.learner).__name__)
-            
-        if self.train_data is not None:
-            train = "Train Data: %i features, %i instances" % \
-                (len(self.train_data.domain), len(self.train_data))
-        elif self.train_data_has_no_class:
-            train = "Train Data has no class variable"
-        elif self.train_data_has_discrete_class:
-            train = "Train Data doesn't have a continuous class"
-            
-        if self.test_data is not None:
-            test = "Test Data: %i features, %i instances" % \
-                (len(self.test_data.domain), len(self.test_data))
-        elif self.train_data:
-            test = "Test data: using training data"
-        
-        self.info_box.setText("\n".join([name, train, test]))
-        
-        if self.learner and self._test_data() is not None:
-            self.commit_if()
-        
-    def invalidate_results(self, which=None):
-        if which is None:
-            self.results = [None for f in self.estimators]
-#            print "Invalidating all"
-        else:
-            for i in which:
-                self.results[i] = None
-#            print "Invalidating", which
-        
-    def run(self):
-        plan = []
-        estimate_index = 0
-        for i, (selected, method, count, offsets) in enumerate(self.estimators):
-            if self.results[i] is None and getattr(self, selected):
-                plan.append((i, method, [estimate_index + offset for offset in offsets]))
-                estimate_index += count
-                
-        estimators = [method() for _, method, _ in plan]
-        
-        if not estimators:
-            return
-            
-        pb = OWGUI.ProgressBar(self, len(self._test_data()))
-        estimates = self.run_estimation(estimators, pb.advance)
-        pb.finish()
-        
-        self.predictions = [v for v, _ in estimates]
-        estimates = [prob.reliability_estimate for _, prob in estimates]
-        
-        for i, (index, method, estimate_indices) in enumerate(plan):
-            self.results[index] = [[e[est_index] for e in estimates] \
-                                   for est_index in estimate_indices]
-        
-    def _test_data(self):
-        if self.test_data is not None:
-            return self.test_data
-        else:
-            return self.train_data
-    
-    def get_estimates(self, estimator, advance=None):
-        test = self._test_data()
-        res = []
-        for i, inst in enumerate(test):
-            value, prob = estimator(inst, result_type=Orange.core.GetBoth)
-            res.append((value, prob))
-            if advance:
-                advance()
-        return res
-                
-    def run_estimation(self, estimators, advance=None):
-        rel = reliability.Learner(self.learner, estimators=estimators)
-        estimator = rel(self.train_data)
-        return self.get_estimates(estimator, advance) 
-    
-    def get_SAVar(self):
-        return reliability.SensitivityAnalysis(e=eval(self.var_e))
-    
-    def get_SABias(self):
-        return reliability.SensitivityAnalysis(e=eval(self.bias_e))
-    
-    def get_BAGV(self):
-        return reliability.BaggingVariance(m=self.bagged_m)
-    
-    def get_LCV(self):
-        return reliability.LocalCrossValidation(k=self.local_cv_k)
-    
-    def get_CNK(self):
-        return reliability.CNeighbours(k=self.local_pe_k)
-    
-    def get_BVCK(self):
-        bagv = reliability.BaggingVariance(m=self.bagged_cn_m)
-        cnk = reliability.CNeighbours(k=self.bagged_cn_k)
-        return reliability.BaggingVarianceCNeighbours(bagv, cnk)
-    
-    def get_Mahalanobis(self):
-        return reliability.Mahalanobis(k=self.mahalanobis_k)
-    
-    def method_selection_changed(self, method=None):
-        self.commit_button.setEnabled(any([getattr(self, selected) \
-                                for selected, _, _, _ in  self.estimators]))
-        self.commit_if()
-    
-    def method_param_changed(self, method=None):
-        if method is not None:
-            self.invalidate_results([method])
-        self.commit_if()
-        
-    def commit_if(self):
-        if self.auto_commit:
-            self.commit()
-        else:
-            self.output_changed = True
-            
-    def commit(self):
-        from Orange import feature as variable
-        name_mapper = {"Mahalanobis absolute": "Mahalanobis"}
-        all_predictions = []
-        all_estimates = []
-        score_vars = []
-        features = []
-        table = None
-        if self.learner and self.train_data is not None \
-                and self._test_data() is not None:
-            self.run()
-            
-            scores = []
-            if self.include_class and not self.include_input_features:
-                original_class = self._test_data().domain.class_var
-                features.append(original_class)
-                
-            if self.include_class:
-                prediction_var = variable.Continuous("Prediction")
-                features.append(prediction_var)
-                
-            if self.include_error:
-                error_var = variable.Continuous("Error")
-                abs_error_var = variable.Continuous("Abs. Error")
-                features.append(error_var)
-                features.append(abs_error_var)
-                
-            for estimates, (selected, method, _, _) in zip(self.results, self.estimators):
-                if estimates is not None and getattr(self, selected):
-                    for estimate in estimates:
-                        name = estimate[0].method_name
-                        name = name_mapper.get(name, name)
-                        var = variable.Continuous(name)
-                        features.append(var)
-                        score_vars.append(var)
-                        all_estimates.append(estimate)
-                    
-            if self.include_input_features:
-                dom = self._test_data().domain
-                attributes = list(dom.attributes) + features
-                domain = Orange.data.Domain(attributes, dom.class_var)
-                domain.add_metas(dom.get_metas())
-                
-                data = Orange.data.Table(domain, self._test_data())
-            else:
-                domain = Orange.data.Domain(features, None)
-                data = Orange.data.Table(domain, [[None] * len(features) for _ in self._test_data()])
-            
-            if self.include_class:
-                for d, inst, pred in zip(data, self._test_data(), self.predictions):
-                    if not self.include_input_features:
-                        d[features[0]] = float(inst.get_class())
-                    d[prediction_var] = float(pred)
-            
-            if self.include_error:
-                for d, inst, pred in zip(data, self._test_data(), self.predictions):
-                    error = float(pred) - float(inst.get_class())
-                    d[error_var] = error
-                    d[abs_error_var] = abs(error)
-                    
-            for estimations, var in zip(all_estimates, score_vars):
-                for d, e in zip(data, estimations):
-                    d[var] = e.estimate
-            
-            table = data
-            
-        self.send("Reliability Scores", table)
-        self.output_changed = False
-        
-        
-if __name__ == "__main__":
-    import sys
-    app = QApplication(sys.argv)
-    w = OWReliability()
-    data = Orange.data.Table("housing")
-    indices = Orange.core.MakeRandomIndices2(p0=20)(data)
-    data = data.select(indices, 0)
-    
-    learner = Orange.regression.tree.TreeLearner()
-    w.set_learner(learner)
-    w.set_train_data(data)
-    w.handleNewSignals()
-    w.show()
-    app.exec_()
-    
-        

File _reliability/widgets/__init__.py

Empty file removed.

File _reliability/widgets/icons/Reliability.svg

-<?xml version="1.0" encoding="utf-8"?>
-<!-- Generator: Adobe Illustrator 16.1.0, SVG Export Plug-In . SVG Version: 6.00 Build 0)  -->
-<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
-<svg version="1.1" id="Layer_1" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" x="0px" y="0px"
-	 width="48px" height="48px" viewBox="0 0 48 48" enable-background="new 0 0 48 48" xml:space="preserve">
-<g>
-	<path fill="#666666" d="M33.414,17.416c0.615,0.877,1.107,1.843,1.461,2.875c0.513,0.219,0.963,0.554,1.316,0.982
-		c-0.37-1.656-1.07-3.186-2.027-4.523C33.945,17.01,33.695,17.232,33.414,17.416z"/>
-	<path fill="#666666" d="M24.009,35.5c-0.003,0-0.006,0-0.009,0c-6.341,0-11.5-5.159-11.5-11.5c0-1.486,0.293-2.903,0.808-4.208
-		c-0.32-0.115-0.617-0.28-0.887-0.48C11.832,20.761,11.5,22.342,11.5,24c0,6.893,5.607,12.5,12.5,12.5
-		c0.052,0,0.103-0.007,0.155-0.008c-0.018-0.062-0.045-0.118-0.06-0.181C24.031,36.042,24.009,35.771,24.009,35.5z"/>
-	<path fill="#666666" d="M35.209,26.536c-0.657,2.905-2.41,5.398-4.808,7.012c0.185,0.275,0.323,0.582,0.426,0.91
-		c3.036-1.989,5.155-5.258,5.586-9.036C36.104,25.892,35.689,26.266,35.209,26.536z"/>
-	<path fill="#666666" d="M17.425,14.579C19.29,13.272,21.555,12.5,24,12.5c1.486,0,2.904,0.293,4.209,0.809
-		c0.081-0.222,0.184-0.438,0.31-0.643c0.054-0.087,0.122-0.159,0.183-0.241C27.248,11.833,25.663,11.5,24,11.5
-		c-2.7,0-5.197,0.87-7.243,2.331C17.012,14.047,17.24,14.296,17.425,14.579z"/>
-	<path fill="#666666" d="M24,31.499c-4.136,0-7.5-3.364-7.5-7.499s3.364-7.499,7.5-7.499c3.119,0,5.797,1.914,6.926,4.628
-		c0.234-0.255,0.512-0.459,0.809-0.634c-1.338-2.94-4.299-4.994-7.734-4.994c-4.687,0-8.5,3.812-8.5,8.499s3.813,8.499,8.5,8.499
-		c3.749,0,6.934-2.443,8.061-5.818c-0.318-0.144-0.611-0.331-0.871-0.56C30.271,29.226,27.398,31.499,24,31.499z"/>
-	<circle fill="#333333" cx="24" cy="24" r="4.5"/>
-	<path fill="#333333" d="M35.931,22.922c-0.109-0.461-0.345-0.858-0.655-1.173C34.813,21.281,34.178,21,33.501,21
-		c-0.192,0-0.387,0.021-0.582,0.068c-0.302,0.072-0.573,0.202-0.816,0.367c-0.339,0.229-0.608,0.535-0.798,0.889
-		c-0.279,0.52-0.385,1.136-0.236,1.755c0.073,0.307,0.203,0.584,0.371,0.83c0.23,0.336,0.535,0.608,0.889,0.797S33.082,26,33.497,26
-		c0.191,0,0.386-0.022,0.58-0.068c0.567-0.135,1.033-0.459,1.37-0.883C35.908,24.469,36.115,23.697,35.931,22.922z"/>
-	<path fill="#333333" d="M16.931,15.922c-0.063-0.267-0.171-0.512-0.308-0.732c-0.179-0.29-0.414-0.533-0.687-0.726
-		C15.522,14.172,15.025,14,14.501,14c-0.192,0-0.387,0.021-0.582,0.068c-1.342,0.32-2.172,1.668-1.851,3.011
-		c0.124,0.523,0.41,0.963,0.786,1.293c0.253,0.222,0.546,0.395,0.867,0.5C13.969,18.953,14.229,19,14.497,19
-		c0.192,0,0.386-0.021,0.581-0.068C16.421,18.611,17.251,17.264,16.931,15.922z"/>
-	<path fill="#333333" d="M29.94,34.995c-0.005-0.024-0.004-0.049-0.01-0.073c-0.075-0.316-0.212-0.601-0.388-0.853
-		C29.08,33.409,28.32,33,27.501,33c-0.192,0-0.387,0.021-0.582,0.068c-1.13,0.27-1.888,1.27-1.909,2.381
-		c-0.004,0.208,0.008,0.418,0.059,0.63c0.029,0.126,0.072,0.245,0.119,0.36C25.57,37.385,26.49,38,27.497,38
-		c0.191,0,0.386-0.022,0.58-0.068C29.396,37.617,30.214,36.312,29.94,34.995z"/>
-	<path fill="#333333" d="M33.931,13.922C33.658,12.773,32.633,12,31.501,12c-0.192,0-0.387,0.021-0.582,0.068
-		c-0.522,0.125-0.964,0.408-1.294,0.784c-0.223,0.253-0.389,0.55-0.494,0.87c-0.139,0.425-0.174,0.89-0.062,1.356
-		C29.342,16.228,30.366,17,31.497,17c0.191,0,0.386-0.021,0.58-0.068c0.267-0.063,0.507-0.178,0.729-0.314
-		c0.29-0.179,0.536-0.41,0.728-0.683C33.933,15.369,34.104,14.647,33.931,13.922z"/>
-	<path fill="#666666" d="M24,7.5C14.902,7.5,7.5,14.902,7.5,24S14.902,40.5,24,40.5c9.098,0,16.5-7.402,16.5-16.5S33.098,7.5,24,7.5
-		z M29.004,38.655c-0.29,0.1-0.584,0.188-0.883,0.271C26.807,39.29,25.429,39.5,24,39.5c-8.547,0-15.5-6.953-15.5-15.5
-		c0-3.272,1.025-6.307,2.763-8.811c0.384-0.554,0.81-1.073,1.26-1.57C15.361,10.484,19.448,8.5,24,8.5
-		c3.271,0,6.303,1.024,8.807,2.761c0.557,0.386,1.079,0.813,1.579,1.267C37.518,15.365,39.5,19.45,39.5,24
-		C39.5,30.794,35.1,36.567,29.004,38.655z"/>
-</g>
-</svg>

File _reliability/widgets/icons/Reliability_60.png

Removed
Old image

File orangecontrib/__init__.py

+# namespace declaration.
+__import__("pkg_resources").declare_namespace(__name__)

File orangecontrib/reliability/__init__.py

+import Orange
+
+import random
+from Orange import statc
+import math
+import warnings
+import numpy
+
+from collections import defaultdict
+from itertools import izip
+
+# Labels and final variables
+labels = ["SAvar", "SAbias", "BAGV", "CNK", "LCV", "BVCK", "Mahalanobis", "ICV"]
+
+"""
+# All the estimators calculation constants
+DO_SA = 0
+DO_BAGV = 1
+DO_CNK = 2
+DO_LCV = 3
+DO_BVCK = 4
+DO_MAHAL = 5
+"""
+
+# All the estimator method constants
+SAVAR_ABSOLUTE = 0
+SABIAS_SIGNED = 1
+SABIAS_ABSOLUTE = 2
+BAGV_ABSOLUTE = 3
+CNK_SIGNED = 4
+CNK_ABSOLUTE = 5
+LCV_ABSOLUTE = 6
+BVCK_ABSOLUTE = 7
+MAHAL_ABSOLUTE = 8
+BLENDING_ABSOLUTE = 9
+ICV_METHOD = 10
+MAHAL_TO_CENTER_ABSOLUTE = 13
+DENS_ABSOLUTE = 14
+ERR_ABSOLUTE = 15
+
+# Type of estimator constant
+SIGNED = 0
+ABSOLUTE = 1
+
+# Names of all the estimator methods
+METHOD_NAME = {0: "SAvar absolute", 1: "SAbias signed", 2: "SAbias absolute",
+               3: "BAGV absolute", 4: "CNK signed", 5: "CNK absolute",
+               6: "LCV absolute", 7: "BVCK_absolute", 8: "Mahalanobis absolute",
+               9: "BLENDING absolute", 10: "ICV", 11: "RF Variance", 12: "RF Std",
+               13: "Mahalanobis to center", 14: "Density based", 15: "Reference expected error"}
+
+select_with_repeat = Orange.core.MakeRandomIndicesMultiple()
+select_with_repeat.random_generator = Orange.misc.Random()
+
+def get_reliability_estimation_list(res, i):
+    return [result.probabilities[0].reliability_estimate[i].estimate for result in res.results], res.results[0].probabilities[0].reliability_estimate[i].signed_or_absolute, res.results[0].probabilities[0].reliability_estimate[i].method
+
+def get_prediction_error_list(res):
+    return [result.actual_class - result.classes[0] for result in res.results]
+
+def get_description_list(res, i):
+    return [result.probabilities[0].reliability_estimate[i].text_description for result in res.results]
+
+def get_pearson_r(res):
+    """
+    :param res: results of evaluation, done using learners,
+        wrapped into :class:`Orange.evaluation.reliability.Classifier`.
+    :type res: :class:`Orange.evaluation.testing.ExperimentResults`
+
+    Return Pearson's coefficient between the prediction error and each of the
+    used reliability estimates. Also, return the p-value of each of
+    the coefficients.
+    """
+    prediction_error = get_prediction_error_list(res)
+    results = []
+    for i in xrange(len(res.results[0].probabilities[0].reliability_estimate)):
+        reliability_estimate, signed_or_absolute, method = get_reliability_estimation_list(res, i)
+        try:
+            if signed_or_absolute == SIGNED:
+                r, p = statc.pearsonr(prediction_error, reliability_estimate)
+            else:
+                r, p = statc.pearsonr([abs(pe) for pe in prediction_error], reliability_estimate)
+        except Exception:
+            r = p = float("NaN")
+        results.append((r, p, signed_or_absolute, method))
+    return results
+
+def get_spearman_r(res):
+    """
+    :param res: results of evaluation, done using learners,
+        wrapped into :class:`Orange.evaluation.reliability.Classifier`.
+    :type res: :class:`Orange.evaluation.testing.ExperimentResults`
+
+    Return Spearman's coefficient between the prediction error and each of the
+    used reliability estimates. Also, return the p-value of each of
+    the coefficients.
+    """
+    prediction_error = get_prediction_error_list(res)
+    results = []
+    for i in xrange(len(res.results[0].probabilities[0].reliability_estimate)):
+        reliability_estimate, signed_or_absolute, method = get_reliability_estimation_list(res, i)
+        try:
+            if signed_or_absolute == SIGNED:
+                r, p = statc.spearmanr(prediction_error, reliability_estimate)
+            else:
+                r, p = statc.spearmanr([abs(pe) for pe in prediction_error], reliability_estimate)
+        except Exception:
+            r = p = float("NaN")
+        results.append((r, p, signed_or_absolute, method))
+    return results
+
+def get_pearson_r_by_iterations(res):
+    """
+    :param res: results of evaluation, done using learners,
+        wrapped into :class:`Orange.evaluation.reliability.Classifier`.
+    :type res: :class:`Orange.evaluation.testing.ExperimentResults`
+
+    Return average Pearson's coefficient over all folds between prediction error
+    and each of the used estimates.
+    """
+    results_by_fold = Orange.evaluation.scoring.split_by_iterations(res)
+    number_of_estimates = len(res.results[0].probabilities[0].reliability_estimate)
+    number_of_instances = len(res.results)
+    number_of_folds = len(results_by_fold)
+    results = [0 for _ in xrange(number_of_estimates)]
+    sig = [0 for _ in xrange(number_of_estimates)]
+    method_list = [0 for _ in xrange(number_of_estimates)]
+
+    for res in results_by_fold:
+        prediction_error = get_prediction_error_list(res)
+        for i in xrange(number_of_estimates):
+            reliability_estimate, signed_or_absolute, method = get_reliability_estimation_list(res, i)
+            try:
+                if signed_or_absolute == SIGNED:
+                    r, _ = statc.pearsonr(prediction_error, reliability_estimate)
+                else:
+                    r, _ = statc.pearsonr([abs(pe) for pe in prediction_error], reliability_estimate)
+            except Exception:
+                r = float("NaN")
+            results[i] += r
+            sig[i] = signed_or_absolute
+            method_list[i] = method
+
+    # Calculate p-values
+    results = [float(res) / number_of_folds for res in results]
+    ps = [p_value_from_r(r, number_of_instances) for r in results]
+
+    return zip(results, ps, sig, method_list)
+
+def p_value_from_r(r, n):
+    """
+    Calculate p-value from the paerson coefficient and the sample size.
+    """
+    df = n - 2
+    t = r * (df / ((-r + 1.0 + 1e-30) * (r + 1.0 + 1e-30))) ** 0.5
+    return statc.betai (df * 0.5, 0.5, df / (df + t * t))
+
+
+# Distances between two discrete probability distributions
+#TODO Document those.
+def normalize_both(p, q):
+    if not p.normalized:
+        p.normalize()
+    if not q.normalized:
+        q.normalize()
+    return p, q
+
+def minkowsky_dist(p, q, m=2):
+    p, q = normalize_both(p, q)
+    dist = 0
+    for i in range(len(p)):
+        dist += abs(p[i]-q[i])**m
+    return dist**(1./m)
+
+def manhattan_distance(p, q):
+    return minkowsky_dist(p, q, m=1)
+
+def euclidean_dist(p, q):
+    return minkowsky_dist(p, q, m=2)
+
+def variance_dist(p, q):
+    return euclidean_dist(p, q) ** 2
+
+def max_dist(p, q):
+    p, q = normalize_both(p, q)
+    return max([abs(p[i]-q[i]) for i in range(len(p))])
+
+def hellinger_dist(p, q):
+    p, q = normalize_both(p, q)
+    dist = 0
+    for i in range(len(p)):
+        dist += (math.sqrt(p[i])-math.sqrt(q[i])) ** 2
+    return dist
+
+def my_log(x):
+    return 0 if x == 0 else x * math.log(x)
+
+def kullback_leibler(p, q):
+    p, q = normalize_both(p, q)
+    dist = 0
+    for i in range(len(p)):
+        dist += my_log(p[i]-q[i])
+    return dist
+
+def cosine(p, q):
+    p, q = normalize_both(p, q)
+    p, q = [pp for pp in p], [qq for qq in q]
+    return 1 - numpy.dot(x,y) / (numpy.linalg.norm(p)*numpy.linalg.norm(q))
+
+
+class Estimate:
+    """
+    Reliability estimate. Contains attributes that describe the results of
+    reliability estimation.
+
+    .. attribute:: estimate
+
+        A numerical reliability estimate.
+
+    .. attribute:: signed_or_absolute
+
+        Determines whether the method used gives a signed or absolute result.
+        Has a value of either :obj:`SIGNED` or :obj:`ABSOLUTE`.
+
+    .. attribute:: method
+
+        An integer ID of reliability estimation method used.
+
+    .. attribute:: method_name
+
+        Name (string) of reliability estimation method used.
+
+    .. attribute:: icv_method
+
+        An integer ID of reliability estimation method that performed best,
+        as determined by ICV, and of which estimate is stored in the
+        :obj:`estimate` field. (:obj:`None` when ICV was not used.)
+
+    .. attribute:: icv_method_name
+
+        Name (string) of reliability estimation method that performed best,
+        as determined by ICV. (:obj:`None` when ICV was not used.)
+
+    """
+    def __init__(self, estimate, signed_or_absolute, method, icv_method= -1):
+        self.estimate = estimate
+        self.signed_or_absolute = signed_or_absolute
+        self.method = method
+        self.method_name = METHOD_NAME[method]
+        self.icv_method = icv_method
+        self.icv_method_name = METHOD_NAME[icv_method] if icv_method != -1 else ""
+        self.text_description = None
+
+class DescriptiveAnalysis:
+    def __init__(self, estimator, desc=["high", "medium", "low"], procentage=[0.00, 0.33, 0.66], name="da"):
+        self.desc = desc
+        self.procentage = procentage
+        self.estimator = estimator
+        self.name = name
+
+    def __call__(self, instances, weight=None, **kwds):
+
+        # Calculate borders using cross validation
+        res = Orange.evaluation.testing.cross_validation([self.estimator], instances)
+        all_borders = []
+        for i in xrange(len(res.results[0].probabilities[0].reliability_estimate)):
+            estimates, signed_or_absolute, method = get_reliability_estimation_list(res, i)
+            sorted_estimates = sorted(abs(x) for x in estimates)
+            borders = [sorted_estimates[int(len(estimates) * p) - 1]  for p in self.procentage]
+            all_borders.append(borders)
+
+        # Learn on whole train data
+        estimator_classifier = self.estimator(instances)
+
+        return DescriptiveAnalysisClassifier(estimator_classifier, all_borders, self.desc)
+
+class DescriptiveAnalysisClassifier:
+    def __init__(self, estimator_classifier, all_borders, desc):
+        self.estimator_classifier = estimator_classifier
+        self.all_borders = all_borders
+        self.desc = desc
+
+    def __call__(self, instance, result_type=Orange.core.GetValue):
+        predicted, probabilities = self.estimator_classifier(instance, Orange.core.GetBoth)
+
+        for borders, estimate in zip(self.all_borders, probabilities.reliability_estimate):
+            estimate.text_description = self.desc[0]
+            for lower_border, text_desc in zip(borders, self.desc):
+                if estimate.estimate >= lower_border:
+                    estimate.text_description = text_desc
+
+        # Return the appropriate type of result
+        if result_type == Orange.core.GetValue:
+            return predicted
+        elif result_type == Orange.core.GetProbabilities:
+            return probabilities
+        else:
+            return predicted, probabilities
+
+class SensitivityAnalysis:
+    """
+    
+    :param e: List of possible :math:`\epsilon` values for SAvar and SAbias
+        reliability estimates.
+    :type e: list of floats
+    
+    :rtype: :class:`Orange.evaluation.reliability.SensitivityAnalysisClassifier`
+    
+    To estimate the reliability of prediction for given instance,
+    the learning set is extended with this instance, labeled with
+    :math:`K + \epsilon (l_{max} - l_{min})`,
+    where :math:`K` denotes the initial prediction,
+    :math:`\epsilon` is sensitivity parameter and :math:`l_{min}` and
+    :math:`l_{max}` denote lower and the upper bound of the learning
+    instances' labels. After computing different sensitivity predictions
+    using different values of :math:`\epsilon`, the prediction are combined
+    into SAvar and SAbias. SAbias can be used in a signed or absolute form.
+
+    :math:`SAvar = \\frac{\sum_{\epsilon \in E}(K_{\epsilon} - K_{-\epsilon})}{|E|}`
+
+    :math:`SAbias = \\frac{\sum_{\epsilon \in E} (K_{\epsilon} - K ) + (K_{-\epsilon} - K)}{2 |E|}`
+    
+    
+    """
+    def __init__(self, e=[0.01, 0.1, 0.5, 1.0, 2.0], name="sa"):
+        self.e = e
+        self.name = name
+
+    def __call__(self, instances, learner):
+        min_value = max_value = instances[0].getclass().value
+        for ex in instances:
+            if ex.getclass().value > max_value:
+                max_value = ex.getclass().value
+            if ex.getclass().value < min_value:
+                min_value = ex.getclass().value
+        return SensitivityAnalysisClassifier(self.e, instances, min_value, max_value, learner)
+
+class SensitivityAnalysisClassifier:
+    def __init__(self, e, instances, min_value, max_value, learner):
+        self.e = e
+        self.instances = instances
+        self.max_value = max_value
+        self.min_value = min_value
+        self.learner = learner
+
+    def __call__(self, instance, predicted, probabilities):
+        # Create new dataset
+        r_data = Orange.data.Table(self.instances)
+
+        # Create new instance
+        modified_instance = Orange.data.Instance(instance)
+
+        # Append it to the data
+        r_data.append(modified_instance)
+
+        # Calculate SAvar & SAbias
+        SAvar = SAbias = 0
+
+        for eps in self.e:
+            # +epsilon
+            r_data[-1].setclass(predicted.value + eps * (self.max_value - self.min_value))
+            c = self.learner(r_data)
+            k_plus = c(instance, Orange.core.GetValue)
+
+            # -epsilon
+            r_data[-1].setclass(predicted.value - eps * (self.max_value - self.min_value))
+            c = self.learner(r_data)
+            k_minus = c(instance, Orange.core.GetValue)
+            #print len(r_data)
+            #print eps*(self.max_value - self.min_value)
+            #print k_plus
+            #print k_minus
+            # calculate part SAvar and SAbias
+            SAvar += k_plus.value - k_minus.value
+            SAbias += k_plus.value + k_minus.value - 2 * predicted.value
+
+        SAvar /= len(self.e)
+        SAbias /= 2 * len(self.e)
+
+        return [Estimate(SAvar, ABSOLUTE, SAVAR_ABSOLUTE),
+                Estimate(SAbias, SIGNED, SABIAS_SIGNED),
+                Estimate(abs(SAbias), ABSOLUTE, SABIAS_ABSOLUTE)]
+
+
+
+class ReferenceExpectedError:
+    """
+
+    :rtype: :class:`Orange.evaluation.reliability.ReferenceExpectedErrorClassifier`
+
+    Reference reliability estimation method for classification as used in Evaluating Reliability of Single
+    Classifications of Neural Networks, Darko Pevec, 2011.
+
+    :math:`O_{ref} = 2 (\hat y - \hat y ^2) = 2 \hat y (1-\hat y)`
+
+    where :math:`\hat y` is the estimated probability of the predicted class.
+
+    Note that for this method, in contrast with all others, a greater estimate means lower reliability (greater
+    expected error).
+
+    """
+    def __init__(self, name="reference"):
+        self.name = name
+
+    def __call__(self, instances, learner):
+        classifier = learner(instances)
+        return ReferenceExpectedErrorClassifier(classifier)
+
+    
+class ReferenceExpectedErrorClassifier:
+
+    def __init__(self, classifier):
+        self.classifier = classifier
+
+    def __call__(self, instance, *args):
+        y_hat = max(self.classifier(instance, Orange.classification.Classifier.GetProbabilities))
+        return [Estimate(2 * y_hat * (1 - y_hat), ABSOLUTE, ERR_ABSOLUTE)]
+
+    
+
+class BaggingVariance:
+    """
+    
+    :param m: Number of bagging models to be used with BAGV estimate
+    :type m: int
+    
+    :rtype: :class:`Orange.evaluation.reliability.BaggingVarianceClassifier`
+    
+    :math:`m` different bagging models are constructed and used to estimate
+    the value of dependent variable for a given instance. In regression,
+    the variance of those predictions is used as a prediction reliability
+    estimate.
+
+    :math:`BAGV = \\frac{1}{m} \sum_{i=1}^{m} (K_i - K)^2`
+
+    where :math:`K = \\frac{\sum_{i=1}^{m} K_i}{m}` and :math:`K_i` are
+    predictions of individual constructed models. Note that a greater value
+    implies greater error.
+
+    For classification, 1 minus the average Euclidean distance between class
+    probability distributions predicted by the model, and distributions
+    predicted by the individual bagged models, is used as the BAGV reliability
+    measure. Note that in this case a greater value implies a better
+    prediction.
+    
+    """
+    def __init__(self, m=50, name="bv"):
+        self.m = m
+        self.name = name
+
+    def __call__(self, instances, learner):
+        classifiers = []
+
+        if instances.domain.class_var.var_type == Orange.feature.Descriptor.Discrete:
+            classifier = learner(instances)
+        else:
+            classifier = None
+
+        # Create bagged classifiers using sampling with replacement
+        for _ in xrange(self.m):
+            selection = select_with_repeat(len(instances))
+            data = instances.select(selection)
+            classifiers.append(learner(data))
+        return BaggingVarianceClassifier(classifiers, classifier)
+
+class BaggingVarianceClassifier: