Ctrl+K

binder

[1]:

import warnings

warnings.filterwarnings("ignore")

Channel Selection in Multivariate Time Series Classification#

Overview#

Sometimes every channel is not required to perform classification; only a few are useful. The [1] proposed a fast channel selection technique for Multivariate Time Classification.

[1] : Fast Channel Selection for Scalable Multivariate Time Series Classification Link

[2]:

from sklearn.linear_model import RidgeClassifierCV
from sklearn.pipeline import make_pipeline

from sktime.datasets import load_UCR_UEA_dataset
from sktime.transformations.panel import channel_selection
from sktime.transformations.panel.rocket import Rocket

1 Initialise the Pipeline#

[3]:

# cs = channel_selection.ElbowClassSum()  # ECS
cs = channel_selection.ElbowClassPairwise()  # ECP

[4]:

rocket_pipeline = make_pipeline(cs, Rocket(), RidgeClassifierCV())

2 Load and Fit the Training Data#

[5]:

data = "BasicMotions"
X_train, y_train = load_UCR_UEA_dataset(data, split="train", return_X_y=True)
X_test, y_test = load_UCR_UEA_dataset(data, split="test", return_X_y=True)

[6]:

rocket_pipeline.fit(X_train, y_train)

[6]:

Pipeline(steps=[('elbowclasspairwise', ElbowClassPairwise()),
                ('rocket', Rocket()),
                ('ridgeclassifiercv',
                 RidgeClassifierCV(alphas=array([ 0.1,  1. , 10. ])))])

3 Classify the Test Data#

[7]:

rocket_pipeline.score(X_test, y_test)

[7]:

1.0

4 Identify channels#

[8]:

rocket_pipeline.steps[0][1].channels_selected_

[8]:

[0, 1]

[9]:

rocket_pipeline.steps[0][1].distance_frame_

[9]:
Centroid_badminton_running Centroid_badminton_standing Centroid_badminton_walking Centroid_running_standing Centroid_running_walking Centroid_standing_walking
0 39.594679 55.752785 48.440779 63.610220 57.247383 10.717044
1 57.681767 24.390543 27.770269 60.458125 62.339120 16.370347
2 20.175911 24.126969 22.331621 25.671979 22.991555 4.897452
3 12.546212 12.439152 12.741854 6.317654 6.695743 3.585273
4 10.101196 8.865871 9.221908 6.520172 6.715702 1.299989
5 23.464251 14.568685 13.953445 18.878429 19.768549 7.228389

5 Standalone#

[10]:

cs.fit(X_train, y_train)

[10]:

ElbowClassPairwise()

6 Distance Matrix#

[11]:

cs.distance_frame_

[11]:
Centroid_badminton_running Centroid_badminton_standing Centroid_badminton_walking Centroid_running_standing Centroid_running_walking Centroid_standing_walking
0 39.594679 55.752785 48.440779 63.610220 57.247383 10.717044
1 57.681767 24.390543 27.770269 60.458125 62.339120 16.370347
2 20.175911 24.126969 22.331621 25.671979 22.991555 4.897452
3 12.546212 12.439152 12.741854 6.317654 6.695743 3.585273
4 10.101196 8.865871 9.221908 6.520172 6.715702 1.299989
5 23.464251 14.568685 13.953445 18.878429 19.768549 7.228389 
[12]:

cs.train_time_

[12]:

13

Generated using nbsphinx. The Jupyter notebook can be found here.

On this page