Time Series Clustering¶
Perform time series clustering in Amazon SageMaker
SageMaker Algorithm |
CPU Training |
GPU Training |
Multi-GPU Training |
Incremental Training |
|---|---|---|---|---|
CNN-KMeans SageMaker Algorithm¶
The CNN-KMeans SageMaker Algorithm performs time series clustering with an unsupervised convolutional neural network (CNN) followed by a K-Means clusterer. The CNN network encodes the input time series into a number of time-independent features, which are then used as input by the K-Means algorithm. The CNN network consists of a stack of exponentially dilated causal convolutional blocks with residual connections and is trained in an unsupervised manner using a contrastive learning procedure that minimizes the triplet loss. The algorithm can be used for time series with different lengths and with missing values. For additional information, see the algorithm's AWS Marketplace listing page and GitHub repository.