-
Notifications
You must be signed in to change notification settings - Fork 68
/
datareduce.py
executable file
·140 lines (127 loc) · 6.45 KB
/
datareduce.py
1
2
3
4
5
6
7
8
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
import numpy as np
import utils
def reduce(x_train, y_train, C, init_clusters_per_class,
clustering_algorithm = 'kmeans', averaging_algorithm = 'dba',
distance_algorithm='dtw'):
"""
Reduces the x_train using an averaging method applied on the result of clustering
:param x_train: The time series to be reduced.
:param y_train: The corresponding labels of the time series to be reduced.
:param C: The number of exemplars per class.
:param init_clusters_per_class: the initial clusters for k-means per class
:param nb_iterations: Nbr of iterations of the clustering algorithm to converge.
:param max_iter: Nbr of iterations of the averaging algorithm.
"""
# get the averaging method (check constants.py)
avg_method = utils.constants.AVERAGING_ALGORITHMS[averaging_algorithm]
# get the clustering algorithm (check constants.py)
cluster_method = utils.constants.CLUSTERING_ALGORITHMS[clustering_algorithm]
# get all unique classes
classes = np.unique(y_train)
# get the number of classes
nb_classes = len(classes)
# initialize list of condensed time series per class
# for each class we have a list of the reduced time series
condensed_data = [[] for i in range(nb_classes)]
# loop through classes
for c in range(nb_classes):
# get time series of current class
c_x_train = x_train[np.where(y_train==c)]
# limit the nb_prototypes
nb_prototypes_per_class = min(C, len(c_x_train))
# cluster the time series
_,affect = cluster_method(c_x_train, nb_prototypes_per_class,
init_clusters = init_clusters_per_class[c],
averaging_algorithm = averaging_algorithm,
distance_algorithm = distance_algorithm)
# loop through all clusters
for i in range(nb_prototypes_per_class):
# get the group of time series affected to cluster index i
data_to_condense = [c_x_train[j] for j, x in enumerate(affect) if x == i]
# check if this cluster center has no affected
if len(data_to_condense) == 0:
# if so do not average
print('Empty cluster ',i, ' for class ',c,' with ',len(c_x_train),' examples')
continue
# condense the data
c_x_train_reduced = avg_method(data_to_condense,
distance_algorithm=distance_algorithm)
# add the average to the list of condensed or reduced data for class c
condensed_data[c].append(c_x_train_reduced)
return reconstruct_reduced_trainset(condensed_data)
# it is the same as reduce but for one class
def reduce_per_class(curr_dir, c_x_train, c_y_train, C, c_init_clusters, class_c,
clustering_algorithm = 'kmeans', averaging_algorithm = 'dba',
distance_algorithm='dtw'):
"""
Reduces the x_train using an averaging method applied on the result of clustering
:param x_train: The time series to be reduced.
:param y_train: The corresponding labels of the time series to be reduced.
:param C: The number of exemplars per class.
:param c_init_clusters: the initial clusters for k-means for the class_c
"""
# get the averaging method (check constants.py)
avg_method = utils.constants.AVERAGING_ALGORITHMS[averaging_algorithm]
# get the clustering algorithm (check constants.py)
cluster_method = utils.constants.CLUSTERING_ALGORITHMS[clustering_algorithm]
# list that contains the prototypes for class_c
condensed_data_c = []
# limit the nb_prototypes
nb_prototypes_per_class = min(C, len(c_x_train))
# cluster the time series
_,affect = cluster_method(curr_dir, c_x_train, nb_prototypes_per_class,
init_clusters = c_init_clusters,
averaging_algorithm = averaging_algorithm,
distance_algorithm = distance_algorithm)
# loop through all clusters
for i in range(nb_prototypes_per_class):
# get the group of time series affected to cluster index i
data_to_condense = [c_x_train[j] for j, x in enumerate(affect) if x == i]
# check if this cluster center has no affected
if len(data_to_condense) == 0:
# if so do not average
print('Empty cluster ',i, ' for class ',class_c,' with ',
len(c_x_train),' examples')
continue
# condense the data
c_x_train_reduced = avg_method(data_to_condense,
distance_algorithm=distance_algorithm)
# add the average to the list of condensed or reduced data for class c
condensed_data_c.append(c_x_train_reduced)
return condensed_data_c
def reconstruct_reduced_trainset(condensed_data):
nb_classes = len(condensed_data)
# construc the reduced_x_train and the reduced_y_train
reduced_x_train = []
reduced_y_train = []
# loop through all the classes
for c in range(nb_classes):
# add the condensed data of the current class to the new x_train
reduced_x_train.extend(condensed_data[c])
C = len(condensed_data[c])
# add the corresponding classes of the condensed data
reduced_y_train.extend([c for i in range(C)])
# return the list of condensed data of all classes
return reduced_x_train, reduced_y_train
def reduce_data_random(x_train, y_train, C):
"""
Reduces the data by randomly selecting C instances from each class
"""
# get all unique classes
classes = np.unique(y_train)
# get the number of classes
nb_classes = len(classes)
# initialize list of condensed time series per class
# for each class we have a list of the reduced time series
condensed_data = [[] for i in range(nb_classes)]
# loop through classes
for c in range(nb_classes):
# get time series of current class
c_x_train = x_train[np.where(y_train==c)]
# limit the nb_prototypes
nb_prototypes_per_class = min(C, len(c_x_train))
# choose C random indices
random_idx = np.random.permutation(len(c_x_train))[:nb_prototypes_per_class]
# get the condensed data based on the random indices
condensed_data[c] = [c_x_train[i] for i in random_idx]
return reconstruct_reduced_trainset(condensed_data)