Tutorial of Time Series Classification

What is Time Series Classification?

Time series classification is a general task that has been applied in many fields. The purpose is to determine which predefined category a time series belongs to by using labeled training data. Time series classification is different from conventional classification because time series data is a sequence with order attribute.

The ECG signal sample shown in the figure below represents a heartbeat activity. It can be distinguished according to different signal characteristics. The left represents normal heartbeat, while the right represents myocardial infarction. (Data from https://timeseriesclassification.com/description.php?Dataset=ECG200）

Classify Windows

It can be seen that we define a time series classification problem as the classification of time series. In the above example, each ECG is a data sample, and each data sample will correspond to a classified label. As shown in the figure below, the input of the model can be n TSDataset.

Classify Windows

PaddleTS provides the components of time series classification, mainly including:

  • The TSDataset data structure is used to adapt the data characteristics of the time series classification scenario, such as the ECG example above.

  • It has implemented two deep learning models for time series classification, including CNN and InceptionTime. It also provides the ClassifyBaseModel, which facilitates developers to build new time series classification models.

  • Accuracy is supported by default to evaluates the time series classification, and also applies the relevant evaluation operator of sklearn.metrics to evaluate the time series classification results.

An example

The following example uses the built-in CNN model to do an time series classification task based on PaddleTS.

1. Prepare data

The built-in API: get_dataset is used to load BasicMotions dataset: BasicMotions.

from paddlets.datasets.repository import get_dataset

ts_x_trains, y_train = get_dataset("BasicMotions_Train")
ts_x_tests, y_test = get_dataset("BasicMotions_Test")

2. Train model

Training time series classification model.

from paddlets.models.classify.dl.cnn import CNNClassifier

network = CNNClassifier(max_epochs=1000, patience=50)
network.fit(ts_x_trains, y_train)

3. Model prediction and evaluation

Use the trained model for prediction and evaluation.

from sklearn.metrics import accuracy_score, f1_score

preds = network.predict(ts_x_tests)
score = accuracy_score(y_test, preds)
f1 = f1_score(y_test, preds, average="macro")

4. Probability Forecasting

Return probability of each class.

preds = network.predict_proba(ts_x_tests)

5. Model persistence

The trained model can be saved to the specified path.

network.save('./model/cnn')

After the model is saved, the model can be loaded for new data prediction.

from paddlets.models.classify.dl.paddle_base import PaddleBaseClassifier

load_network = PaddleBaseClassifier.load('./model/cnn')