-
Notifications
You must be signed in to change notification settings - Fork 89
/
tapnet.py
266 lines (228 loc) · 8.64 KB
/
tapnet.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
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
# -*- coding: utf-8 -*-
"""Time Convolutional Neural Network (CNN) for classification."""
__author__ = [
"Jack Russon",
]
__all__ = [
"TapNetRegressor",
]
import gc
from copy import deepcopy
from sklearn.utils import check_random_state
from aeon.networks.tapnet import TapNetNetwork
from aeon.regression.deep_learning.base import BaseDeepRegressor
from aeon.utils.validation._dependencies import _check_dl_dependencies
class TapNetRegressor(BaseDeepRegressor):
"""Time series attentional prototype network (TapNet).
Adapted from the implementation used in [1]_.
TapNet was initially proposed for multivariate time series classification. The is an
adaptation for time series regression. TapNet comprises these components: random
dimension permutation, multivariate time series encoding, and attentional
prototype learning.
Parameters
----------
filter_sizes : array of int, default = (256, 256, 128)
sets the kernel size argument for each convolutional block.
Controls number of convolutional filters
and number of neurons in attention dense layers.
kernel_size : array of int, default = (8, 5, 3)
controls the size of the convolutional kernels
layers : array of int, default = (500, 300)
size of dense layers
n_epochs : int, default = 2000
number of epochs to train the model
batch_size : int, default = 16
number of samples per update
dropout : float, default = 0.5
dropout rate, in the range [0, 1)
dilation : int, default = 1
dilation value
activation : str, default = "sigmoid"
activation function for the last output layer
loss : str, default = "mean_squared_error"
loss function for the classifier
optimizer : str or None, default = "Adam(lr=0.01)"
gradient updating function for the classifier
use_bias : bool, default = True
whether to use bias in the output dense layer
use_rp : bool, default = True
whether to use random projections
use_att : bool, default = True
whether to use self attention
use_lstm : bool, default = True
whether to use an LSTM layer
use_cnn : bool, default = True
whether to use a CNN layer
verbose : bool, default = False
whether to output extra information
random_state : int or None, default = None
seed for random
References
----------
.. [1] Zhang et al. Tapnet: Multivariate time series classification with
attentional prototypical network,
Proceedings of the AAAI Conference on Artificial Intelligence
34(4), 6845-6852, 2020
Notes
-----
The Implementation of TapNet found at https://github.com/kdd2019-tapnet/tapnet
Currently does not implement custom distance matrix loss function
or class based self attention.
"""
_tags = {"python_dependencies": "tensorflow"}
def __init__(
self,
n_epochs=2000,
batch_size=16,
dropout=0.5,
filter_sizes=(256, 256, 128),
kernel_size=(8, 5, 3),
dilation=1,
layers=(500, 300),
use_rp=True,
activation=None,
rp_params=(-1, 3),
use_bias=True,
use_att=True,
use_lstm=True,
use_cnn=True,
random_state=None,
padding="same",
loss="mean_squared_error",
optimizer=None,
metrics=None,
callbacks=None,
verbose=False,
):
_check_dl_dependencies(severity="error")
super(TapNetRegressor, self).__init__()
self.batch_size = batch_size
self.random_state = random_state
self.kernel_size = kernel_size
self.layers = layers
self.rp_params = rp_params
self.filter_sizes = filter_sizes
self.activation = activation
self.use_att = use_att
self.use_bias = use_bias
self.dilation = dilation
self.padding = padding
self.n_epochs = n_epochs
self.loss = loss
self.optimizer = optimizer
self.metrics = metrics
self.callbacks = callbacks
self.verbose = verbose
self.dropout = dropout
self.use_lstm = use_lstm
self.use_cnn = use_cnn
# parameters for random projection
self.use_rp = use_rp
self.rp_params = rp_params
self._network = TapNetNetwork(
dropout=self.dropout,
filter_sizes=self.filter_sizes,
kernel_size=self.kernel_size,
dilation=self.dilation,
layers=self.layers,
use_rp=self.use_rp,
rp_params=self.rp_params,
use_att=self.use_att,
use_lstm=self.use_lstm,
use_cnn=self.use_cnn,
random_state=self.random_state,
padding=self.padding,
)
def build_model(self, input_shape, **kwargs):
"""Construct a complied, un-trained, keras model that is ready for training.
In aeon, time series are stored in numpy arrays of shape (d,m), where d
is the number of dimensions, m is the series length. Keras/tensorflow assume
data is in shape (m,d). This method also assumes (m,d). Transpose should
happen in fit.
Parameters
----------
input_shape : tuple
The shape of the data fed into the input layer, should be (m, d)
Returns
-------
output: a compiled Keras model
"""
import tensorflow as tf
from tensorflow import keras
tf.random.set_seed(self.random_state)
metrics = ["mean_squared_error"] if self.metrics is None else self.metrics
input_layer, output_layer = self._network.build_network(input_shape, **kwargs)
output_layer = keras.layers.Dense(
units=1, activation=self.activation, use_bias=self.use_bias
)(output_layer)
self.optimizer_ = (
keras.optimizers.Adam(learning_rate=0.01)
if self.optimizer is None
else self.optimizer
)
model = keras.models.Model(inputs=input_layer, outputs=output_layer)
model.compile(
loss=self.loss,
optimizer=self.optimizer_,
metrics=metrics,
)
return model
def _fit(self, X, y):
"""
Fit the regressor on the training set (X, y).
Parameters
----------
X : np.ndarray of shape = (n_instances(n), n_dimensions(d), series_length(m))
Input training samples
y : np.ndarray of shape n
Input training responses
Returns
-------
self: object
"""
# Transpose to conform to expectation format from keras
X = X.transpose(0, 2, 1)
check_random_state(self.random_state)
self.input_shape = X.shape[1:]
self.model_ = self.build_model(self.input_shape)
if self.verbose:
self.model_.summary()
self.history = self.model_.fit(
X,
y,
batch_size=self.batch_size,
epochs=self.n_epochs,
verbose=self.verbose,
callbacks=deepcopy(self.callbacks) if self.callbacks else [],
)
gc.collect()
return self
@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.
For classifiers, a "default" set of parameters should be provided for
general testing, and a "results_comparison" set for comparing against
previously recorded results if the general set does not produce suitable
probabilities to compare against.
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`.
"""
param1 = {
"n_epochs": 10,
"batch_size": 4,
"padding": "valid",
"filter_sizes": (16, 16, 16),
"kernel_size": (3, 3, 1),
"layers": (25, 50),
}
return [param1]