Demo of SAST transform

Credit

This demo is a copy of the ROCKET demo, where SAST replaces ROCKET. Since SAST does not handle multivariate time series yet, multivariate sections are removed

Overview

SAST [1]_ first randomly selects k time series from each class (they are called reference time series). Then SAST generates all the subsequences of the specified lengths from these reference time series. These subsequences are then used to transform a time series dataset, replacing each time series by the vector of its distance to each subsequence.

[1] Mbouopda, Michael Franklin, and Engelbert Mephu Nguifo. “Scalable and accurate subsequence transform for time series classification.” Pattern Recognition 147 (2023): 110121, https://www.sciencedirect.com/science/article/abs/pii/S003132032300818X, https://uca.hal.science/hal-03087686/document

---

Contents

1. Imports

2. Univariate Time Series

3. Pipeline Example

4. SASTClassifier Example

---

1 Imports

Import example data, SAST, and a classifier (RidgeClassifierCV from scikit-learn), as well as NumPy and make_pipeline from scikit-learn.

Note: SAST compiles (via Numba) on import, which may take a few seconds.

import numpy as np
from sklearn.linear_model import RidgeClassifierCV
from sklearn.pipeline import make_pipeline

from aeon.classification.shapelet_based import SASTClassifier
from aeon.datasets import load_classification
from aeon.transformations.collection.shapelet_based import SAST

2 Univariate Time Series

We can transform the data using SAST and separately fit a classifier, or we can use SAST together with a classifier in a pipeline (section 4, below).

2.1 Load the Training Data

For more details on the data set, see the univariate time series classification notebook.

X_train, y_train = load_classification("Chinatown", split="train")

2.2 Initialise SAST and Transform the Training Data

sast = SAST()
sast.fit(X_train, y_train)
X_train_transform = sast.transform(X_train)

OMP: Info #276: omp_set_nested routine deprecated, please use omp_set_max_active_levels instead.

2.3 Fit a Classifier

We recommend using RidgeClassifierCV from scikit-learn for smaller datasets (fewer than approx. 20K training examples), and using logistic regression trained using stochastic gradient descent for larger datasets.

classifier = RidgeClassifierCV(alphas=np.logspace(-3, 3, 10))
classifier.fit(X_train_transform, y_train)

RidgeClassifierCV(alphas=array([1.00000000e-03, 4.64158883e-03, 2.15443469e-02, 1.00000000e-01,
       4.64158883e-01, 2.15443469e+00, 1.00000000e+01, 4.64158883e+01,
       2.15443469e+02, 1.00000000e+03]))

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

2.4 Load and Transform the Test Data

X_test, y_test = load_classification("Chinatown", split="test")
X_test_transform = sast.transform(X_test)

2.5 Classify the Test Data

classifier.score(X_test_transform, y_test)

0.9795918367346939

---

3 Pipeline Example

We can use SAST together with RidgeClassifierCV (or another classifier) in a pipeline. We can then use the pipeline like a self-contained classifier, with a single call to fit, and without having to separately transform the data, etc.

3.1 Initialise the Pipeline

sast_pipeline = make_pipeline(SAST(), RidgeClassifierCV(alphas=np.logspace(-3, 3, 10)))

3.2 Load and Fit the Training Data

X_train, y_train = load_classification("Chinatown", split="train")

# it is necessary to pass y_train to the pipeline
# y_train is not used for the transform, but it is used by the classifier
sast_pipeline.fit(X_train, y_train)

Pipeline(steps=[('sast', SAST()),
                ('ridgeclassifiercv',
                 RidgeClassifierCV(alphas=array([1.00000000e-03, 4.64158883e-03, 2.15443469e-02, 1.00000000e-01,
       4.64158883e-01, 2.15443469e+00, 1.00000000e+01, 4.64158883e+01,
       2.15443469e+02, 1.00000000e+03])))])

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

3.3 Load and Classify the Test Data

X_test, y_test = load_classification("Chinatown", split="test")

sast_pipeline.score(X_test, y_test)

0.956268221574344

4 SASTClassifier Example

This is a wrapper for building a pipeline seamelessly using SAST transformer

4.1 Ininitialize the classifier

clf = SASTClassifier(seed=42)
clf

SASTClassifier(seed=42)

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

4.2 Fit the classifier on the training data

clf.fit(X_train, y_train)

SASTClassifier(seed=42)

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

4.3 Classify the test data

clf.score(X_test, y_test)

0.9650145772594753

4.4 Interpretability

SASTClassifier’s interpretability can be achieved by visualizing the most important subsequences used for the transformation.

The importance of features can be computed by different means. In this demo, we considere the absolute values of the weights of the ridge classifier as the feature importances.

4.4.1 Interpretability of class 1

fig = clf.plot_most_important_feature_on_ts(
    X_test[y_test == "1"][0, 0], clf._classifier.coef_[0]
)

4.4.2 Interpretability of class 2

fig = clf.plot_most_important_feature_on_ts(
    X_test[y_test == "2"][0, 0], clf._classifier.coef_[0]
)

4.4.3 Interpretability summary

We can see that the most important subsequences matche the instance from class 1 quite exactly, meaning that these subsequences are characteristic of this class. On the contrary, they failed to match the instance from class 2. These subsequences can be called Shapelets for this dataset.

Feel free to play with different time series from the dataset, or with a different dataset

---

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