-
Notifications
You must be signed in to change notification settings - Fork 89
/
_time_series_neighbors.py
247 lines (205 loc) · 8.93 KB
/
_time_series_neighbors.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
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
234
235
236
237
238
239
240
241
242
243
244
245
246
247
# -*- coding: utf-8 -*-
"""KNN time series classification.
A KNN classifier which supports time series distance measures.
The class has hardcoded string references to numba based distances in aeon.distances.
It can also be used with callable distance functions.
"""
__author__ = ["TonyBagnall", "GuiArcencio"]
__all__ = ["KNeighborsTimeSeriesClassifier"]
import numpy as np
from aeon.classification.base import BaseClassifier
from aeon.distances import get_distance_function
WEIGHTS_SUPPORTED = ["uniform", "distance"]
class KNeighborsTimeSeriesClassifier(BaseClassifier):
"""
K-Nearest Neighbour Time Series Classifier.
A KNN classifier which supports time series distance measures.
It has hardcoded string references to numba based distances in aeon.distances,
and can also be used with callables, or aeon (pairwise transformer) estimators.
Parameters
----------
n_neighbors : int, default =1
k for knn.
weights : str or callable, default = 'uniform'
Mechanism for weighting a vote one of: 'uniform', 'distance', or a callable
function.
distance : str or callable, default ='dtw'
Distance measure between time series.
if str, must be one of the following strings:
'euclidean', 'squared', 'dtw', 'ddtw', 'wdtw', 'wddtw',
'lcss', 'edr', 'erp', 'msm', 'twe', 'mpdist'
this will substitute a hard-coded distance metric from aeon.distances
When mpdist is used, the subsequence length (parameter m) must be set
Example: knn_mpdist = KNeighborsTimeSeriesClassifier(
distance='mpdist', distance_params={'m':30})
if callable, must be of signature (X: np.ndarray, X2: np.ndarray) -> np.ndarray
distance_params : dict, default = None
Dictionary for metric parameters for the case that distance is a str.
n_jobs : int, default = None
The number of parallel jobs to run for neighbors search.
``None`` means 1 unless in a :obj:`joblib.parallel_backend` context.
``-1`` means using all processors. See :term:`Glossary <n_jobs>`
for more details.
Parameter for compatibility purposes, still unimplemented.
Examples
--------
>>> from aeon.datasets import load_unit_test
>>> from aeon.classification.distance_based import KNeighborsTimeSeriesClassifier
>>> X_train, y_train = load_unit_test(split="train")
>>> X_test, y_test = load_unit_test(split="test")
>>> classifier = KNeighborsTimeSeriesClassifier(distance="euclidean")
>>> classifier.fit(X_train, y_train)
KNeighborsTimeSeriesClassifier(...)
>>> y_pred = classifier.predict(X_test)
"""
_tags = {
"capability:multivariate": True,
"capability:unequal_length": True,
"X_inner_mtype": ["np-list", "numpy3D"],
"algorithm_type": "distance",
}
def __init__(
self,
distance="dtw",
distance_params=None,
n_neighbors=1,
weights="uniform",
n_jobs=1,
):
self.distance = distance
self.distance_params = distance_params
self.n_neighbors = n_neighbors
self.n_jobs = n_jobs
self._distance_params = distance_params
if self._distance_params is None:
self._distance_params = {}
if weights not in WEIGHTS_SUPPORTED:
raise ValueError(
f"Unrecognised kNN weights: {weights}. "
f"Allowed values are: {WEIGHTS_SUPPORTED}. "
)
self.weights = weights
super(KNeighborsTimeSeriesClassifier, self).__init__()
def _fit(self, X, y):
"""Fit the model using X as training data and y as target values.
Parameters
----------
X : 3D np.ndarray of shape = (n_cases, n_channels, n_timepoints) or list of
shape [n_cases] of 2D arrays shape (n_channels,n_timepoints_i)
If the series are all equal length, a numpy3D will be passed. If unequal,
a list of 2D numpy arrays is passed, which may have different lengths.
y : array-like, shape = (n_cases)
The class labels.
"""
if isinstance(self.distance, str):
self.metric_ = get_distance_function(metric=self.distance)
self.X_ = X
self.classes_, self.y_ = np.unique(y, return_inverse=True)
return self
def _predict_proba(self, X):
"""Return probability estimates for the provided data.
Parameters
----------
X : 3D np.ndarray of shape = (n_cases, n_channels, n_timepoints) or list of
shape[n_cases] of 2D arrays shape (n_channels,n_timepoints_i)
If the series are all equal length, a numpy3D will be passed. If
unequal, a list of 2D numpy arrays is passed, which may have
different lengths.
Returns
-------
p : array of shape = (n_cases, n_classes_)
The class probabilities of the input samples. Classes are ordered
by lexicographic order.
"""
self.check_is_fitted()
preds = np.zeros((len(X), len(self.classes_)))
for i in range(len(X)):
idx, weights = self._kneighbors(X[i])
for id, w in zip(idx, weights):
predicted_class = self.y_[id]
preds[i, predicted_class] += w
preds[i] = preds[i] / np.sum(preds[i])
return preds
def _predict(self, X):
"""Predict the class labels for the provided data.
Parameters
----------
X : 3D np.ndarray of shape = (n_cases, n_channels, n_timepoints) or list of
shape[n_cases] of 2D arrays shape (n_channels,n_timepoints_i)
If the series are all equal length, a numpy3D will be passed. If
unequal, a list of 2D numpy arrays is passed, which may have
different lengths.
Returns
-------
y : array of shape (n_cases)
Class labels for each data sample.
"""
self.check_is_fitted()
preds = np.empty(len(X), dtype=self.classes_.dtype)
for i in range(len(X)):
scores = np.zeros(len(self.classes_))
idx, weights = self._kneighbors(X[i])
for id, w in zip(idx, weights):
predicted_class = self.y_[id]
scores[predicted_class] += w
preds[i] = self.classes_[np.argmax(scores)]
return preds
def _kneighbors(self, X):
"""Find the K-neighbors of a point.
Returns indices and weights of each point.
Parameters
----------
X : 3D np.ndarray of shape = (n_cases, n_channels, n_timepoints) or list of
shape[n_cases] of 2D arrays shape (n_channels,n_timepoints_i)
If the series are all equal length, a numpy3D will be passed. If
unequal, a list of 2D numpy arrays is passed, which may have
different lengths.
Returns
-------
ind : array
Indices of the nearest points in the population matrix.
ws : array
Array representing the weights of each neighbor.
"""
distances = np.array(
[
self.metric_(X, self.X_[j], **self._distance_params)
for j in range(len(self.X_))
]
)
# Find indices of k nearest neighbors using partitioning:
# [0..k-1], [k], [k+1..n-1]
# They might not be ordered within themselves,
# but it is not necessary and partitioning is
# O(n) while sorting is O(nlogn)
closest_idx = np.argpartition(distances, self.n_neighbors)
closest_idx = closest_idx[: self.n_neighbors]
if self.weights == "distance":
ws = distances[closest_idx]
ws = ws**2
# Using epsilon ~= 0 to avoid division by zero
ws = 1 / (ws + np.finfo(float).eps)
elif self.weights == "uniform":
ws = np.repeat(1.0, self.n_neighbors)
else:
raise Exception(f"Invalid kNN weights: {self.weights}")
return closest_idx, ws
@classmethod
def get_test_params(cls, parameter_set="default"):
"""Return testing parameter settings for the estimator.
Parameters
----------
parameter_set : str, default="default"
Name of the set of test parameters to return, for use in tests. If no
special parameters are defined for a value, will return `"default"` set.
Returns
-------
params : dict or list of dict, default={}
Parameters to create testing instances of the class.
Each dict are parameters to construct an "interesting" test instance, i.e.,
`MyClass(**params)` or `MyClass(**params[i])` creates a valid test instance.
`create_test_instance` uses the first (or only) dictionary in `params`.
"""
# non-default distance and algorithm
params1 = {"distance": "euclidean"}
return [params1]