Skip to content

mcikmeans

MCIKMeans

Implementation of K-Means with Minimization of Cluster Impurity (MCI-Kmeans), as described in [1].

This algorithm implements a semi-supervised version of K-Means, that aims to minimize the intra-cluster dispersion while also minimizing the impurity of each cluster.

[1] Masud, Mohammad M., et al. "A practical approach to classify evolving data streams: Training with limited amount of labeled data." 2008 Eighth IEEE International Conference on Data Mining. IEEE, 2008.

Parameters:

Name Type Description Default
n_clusters int

Number of clusters to generate

 8
max_iter int

Maximum number of iterations of the M-Step

 300
conditional_mode_max_iter int

Maximum number of iterations of the E-Step

 300
random_state int

Seed for the random number generation. Makes the algorithm deterministic if a number is provided.

 None

Attributes:

Name Type Description
clusters dict

Dictionary containing each cluster with their label as key
cluster_centers_ ndarray

Array containing the coordinates of the cluster centers
labels_ ndarray

Labels of each point
Source code in streamndr/utils/mcikmeans.py 
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
class MCIKMeans():
    """Implementation of K-Means with Minimization of Cluster Impurity (MCI-Kmeans), as described in [1].

    This algorithm implements a semi-supervised version of K-Means, that aims to minimize the intra-cluster dispersion while also minimizing the impurity of each cluster.

    [1] Masud, Mohammad M., et al. "A practical approach to classify evolving data streams: Training with limited amount of labeled data." 
    2008 Eighth IEEE International Conference on Data Mining. IEEE, 2008.

    Parameters
    ----------
    n_clusters : int
        Number of clusters to generate
    max_iter : int
        Maximum number of iterations of the M-Step
    conditional_mode_max_iter : int
        Maximum number of iterations of the E-Step
    random_state : int
        Seed for the random number generation. Makes the algorithm deterministic if a number is provided.

    Attributes
    ----------
    clusters : dict
        Dictionary containing each cluster with their label as key
    cluster_centers_ : numpy.ndarray
        Array containing the coordinates of the cluster centers
    labels_ : numpy.ndarray
        Labels of each point
    """
    def __init__(self,
                 n_clusters=8,
                 max_iter=300,
                 conditional_mode_max_iter=300,
                 random_state=None):

        self.n_clusters = n_clusters
        self.max_iter = max_iter
        self.conditional_mode_max_iter = conditional_mode_max_iter
        self.random_state = random_state

        if random_state is not None:
            random.seed(random_state)

        self.clusters = []

    def fit(self, X, y):
        """Compute MCI-Kmeans clustering.

        Parameters
        ----------
        X : numpy.ndarray
            Samples
        y : list of int
            Labels of the samples, expects -1 if the label is not known

        Returns
        -------
        MCIKmeans
            Fitted estimator
        """
        y = np.array(y)

        samples_per_class = {}
        number_of_centroids = {}
        unlabeled_samples = []

        nb_labeled_samples = len(X[y!=-1])
        if nb_labeled_samples == 0:
            raise ValueError("MCIKMeans requires at least one labeled sample (y != -1).")
        remaining = 0
        for label in np.unique(y):
            if label == -1:
                unlabeled_samples = X[y==-1]
                continue

            samples_per_class[label] = X[y==label]

            weight = self.n_clusters * len(samples_per_class[label]) / nb_labeled_samples
            remaining += weight - round(weight)
            number_of_centroids[label] = round(weight)

        while(remaining > 0):
            #Find the label with the smallest number of centroids and add the remainder to it
            key_with_min_value = min(number_of_centroids, key=lambda i: number_of_centroids[i])
            number_of_centroids[key_with_min_value] += 1

            remaining -= 1

        centroids = []
        for label in samples_per_class:
            class_centroids = self._init_centroids(samples_per_class[label], number_of_centroids[label])
            centroids.extend(class_centroids)

            if (len(class_centroids) < number_of_centroids[label]) and (len(unlabeled_samples) > 0):
                filling_samples = copy.deepcopy(unlabeled_samples)

                while ((len(class_centroids) < number_of_centroids[label]) and (len(filling_samples) > 0)):
                    choice = filling_samples.pop(random.randrange(len(filling_samples)))
                    centroids.append(choice)
                    class_centroids.append(choice)

        for i in range(len(centroids)):
            self.clusters.append(ImpurityBasedCluster(i, centroids[i]))

        iterations = 0
        changing = True

        while changing and iterations < self.max_iter:
            changing = self._iterative_conditional_mode(samples_per_class, unlabeled_samples)

            for cluster in self.clusters:
                if cluster.n > 0:
                    cluster.update_properties()

            iterations += 1

        self.cluster_centers_ = np.array([cluster.centroid for cluster in self.clusters])
        self.labels_ = self.predict(X)

        return self


    def predict(self, X):
        """Predict the closest cluster each sample in X belongs to.

        Parameters
        ----------
        X : numpy.ndarray
            Samples to predict

        Returns
        -------
        numpy.ndarray
            Index of the cluster each sample belongs to
        """
        from streamndr.utils.cluster_utils import get_closest_clusters
        labels, _ = get_closest_clusters(X, [cluster.centroid for cluster in self.clusters])

        return labels

    def fit_predict(self, X, y):
        """Compute cluster centers and predict cluster index for each sample. Convenience method; equivalent to calling fit(X) followed by predict(X).

        Parameters
        ----------
        X : numpy.ndarray
            Samples
        y : list of int
            Labels of the samples, expects -1 if the label is not known

        Returns
        -------
        numpy.ndarray
            Index of the cluster each sample belongs to
        """
        return self.fit(X, y).labels_

    def _init_centroids(self, samples, numbers_of_centroids):
        centroids = []

        if len(samples) <= numbers_of_centroids:
            centroids.extend(samples)
            return centroids

        candidates = copy.deepcopy(samples).tolist()

        for i in range(numbers_of_centroids):
            selected = candidates.pop(random.randrange(len(candidates)))

            centroids.append(selected)

        return centroids

    def _iterative_conditional_mode(self, samples_per_class, unlabeled_samples):

        _labeled_samples = []
        for key, value in samples_per_class.items():
            _labeled_samples.extend([ShortMemInstance(x, None, key) for x in value])

        _unlabeled_samples = [ShortMemInstance(x, None, -1) for x in unlabeled_samples]

        iterations = 0
        changed = True
        no_change = True

        while iterations < self.conditional_mode_max_iter and changed:
            total_nb_samples = len(_labeled_samples) + len(_unlabeled_samples)

            iterations += 1
            changed = False

            for i in range(total_nb_samples):
                sample = None

                if (len(_labeled_samples) > 0) and ((len(_unlabeled_samples) == 0) or bool(random.getrandbits(1))):
                    sample = _labeled_samples.pop(random.randrange(len(_labeled_samples)))

                else:
                    sample = _unlabeled_samples.pop(random.randrange(len(_unlabeled_samples)))

                previous_cluster_id = sample.timestamp

                if previous_cluster_id is not None:
                    self.clusters[previous_cluster_id].remove_sample(sample)
                    sample.timestamp = None

                distances = np.linalg.norm([cluster.centroid for cluster in self.clusters] - sample.point, axis=1)

                if sample.y_true != -1:
                    entropies = np.array([cluster.entropy for cluster in self.clusters])
                    dissimilarities = np.array([cluster.dissimilarity_count(sample) for cluster in self.clusters])
                    distances = distances * (1 + entropies * dissimilarities)

                chosen_cluster = np.argmin(distances)


                self.clusters[chosen_cluster].add_sample(sample)
                sample.timestamp = chosen_cluster

                self.clusters[chosen_cluster].update_entropy()

                if self.clusters[chosen_cluster].label != previous_cluster_id:
                    changed = True
                    no_change = False

        return not no_change

fit(X, y)

Compute MCI-Kmeans clustering.

Parameters:

Name Type Description Default
X ndarray

Samples

 required
y list of int

Labels of the samples, expects -1 if the label is not known

 required

Returns:

Type Description
MCIKmeans

Fitted estimator
Source code in streamndr/utils/mcikmeans.py 
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
def fit(self, X, y):
    """Compute MCI-Kmeans clustering.

    Parameters
    ----------
    X : numpy.ndarray
        Samples
    y : list of int
        Labels of the samples, expects -1 if the label is not known

    Returns
    -------
    MCIKmeans
        Fitted estimator
    """
    y = np.array(y)

    samples_per_class = {}
    number_of_centroids = {}
    unlabeled_samples = []

    nb_labeled_samples = len(X[y!=-1])
    if nb_labeled_samples == 0:
        raise ValueError("MCIKMeans requires at least one labeled sample (y != -1).")
    remaining = 0
    for label in np.unique(y):
        if label == -1:
            unlabeled_samples = X[y==-1]
            continue

        samples_per_class[label] = X[y==label]

        weight = self.n_clusters * len(samples_per_class[label]) / nb_labeled_samples
        remaining += weight - round(weight)
        number_of_centroids[label] = round(weight)

    while(remaining > 0):
        #Find the label with the smallest number of centroids and add the remainder to it
        key_with_min_value = min(number_of_centroids, key=lambda i: number_of_centroids[i])
        number_of_centroids[key_with_min_value] += 1

        remaining -= 1

    centroids = []
    for label in samples_per_class:
        class_centroids = self._init_centroids(samples_per_class[label], number_of_centroids[label])
        centroids.extend(class_centroids)

        if (len(class_centroids) < number_of_centroids[label]) and (len(unlabeled_samples) > 0):
            filling_samples = copy.deepcopy(unlabeled_samples)

            while ((len(class_centroids) < number_of_centroids[label]) and (len(filling_samples) > 0)):
                choice = filling_samples.pop(random.randrange(len(filling_samples)))
                centroids.append(choice)
                class_centroids.append(choice)

    for i in range(len(centroids)):
        self.clusters.append(ImpurityBasedCluster(i, centroids[i]))

    iterations = 0
    changing = True

    while changing and iterations < self.max_iter:
        changing = self._iterative_conditional_mode(samples_per_class, unlabeled_samples)

        for cluster in self.clusters:
            if cluster.n > 0:
                cluster.update_properties()

        iterations += 1

    self.cluster_centers_ = np.array([cluster.centroid for cluster in self.clusters])
    self.labels_ = self.predict(X)

    return self

fit_predict(X, y)

Compute cluster centers and predict cluster index for each sample. Convenience method; equivalent to calling fit(X) followed by predict(X).

Parameters:

Name Type Description Default
X ndarray

Samples

 required
y list of int

Labels of the samples, expects -1 if the label is not known

 required

Returns:

Type Description
ndarray

Index of the cluster each sample belongs to
Source code in streamndr/utils/mcikmeans.py 
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
def fit_predict(self, X, y):
    """Compute cluster centers and predict cluster index for each sample. Convenience method; equivalent to calling fit(X) followed by predict(X).

    Parameters
    ----------
    X : numpy.ndarray
        Samples
    y : list of int
        Labels of the samples, expects -1 if the label is not known

    Returns
    -------
    numpy.ndarray
        Index of the cluster each sample belongs to
    """
    return self.fit(X, y).labels_

predict(X)

Predict the closest cluster each sample in X belongs to.

Parameters:

Name Type Description Default
X ndarray

Samples to predict

 required

Returns:

Type Description
ndarray

Index of the cluster each sample belongs to
Source code in streamndr/utils/mcikmeans.py 
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
def predict(self, X):
    """Predict the closest cluster each sample in X belongs to.

    Parameters
    ----------
    X : numpy.ndarray
        Samples to predict

    Returns
    -------
    numpy.ndarray
        Index of the cluster each sample belongs to
    """
    from streamndr.utils.cluster_utils import get_closest_clusters
    labels, _ = get_closest_clusters(X, [cluster.centroid for cluster in self.clusters])

    return labels