Time2Feat: Learning Interpretable Representations for Multivariate Time Series Clustering

Time2Feat is an end-to-end machine learning system for multivariate time series clustering. The system is the first to leverage both inter-signal and intra-signal features of the time series. While relying on state-of-the-art feature extraction approaches allows to further refine the features by choosing the most appropriate ones and incorporating human feedback in the feature selection process

For a detailed description of the work please read our paper. Please cite the paper if you use the code from this repository in your work.

@article{DBLP:journals/pvldb/BonifatiB0T22,
  author       = {Angela Bonifati and
                  Francesco Del Buono and
                  Francesco Guerra and
                  Donato Tiano},
  title        = {Time2Feat: Learning Interpretable Representations for Multivariate
                  Time Series Clustering},
  journal      = {Proc. {VLDB} Endow.},
  volume       = {16},
  number       = {2},
  pages        = {193--201},
  year         = {2022}
}

Installation

time2feat was tested on Python 3.7 on Linux amd Windows machines. It is recommended to use a virtual environment ( See: python3 venv doc).

Clone the project into local and install time2feat package:

// Virtual environment creation
$ cd source
$ virtualenv -p python3 venv
$ source venv/bin/activate

// Dependecy installation
$ pip install -r requirements.txt

Quick Start

Get started with time2feat

import numpy as np
from t2f.extraction.extractor import feature_extraction
from t2f.utils.importance_old import feature_selection
from t2f.model.clustering import ClusterWrapper

# 10 multivariate time series with 100 timestamps and 3 signals each
arr = np.random.randn(10, 100, 3)
arr[5:] = arr[5:] * 100

labels = {}  # unsupervised mode
# labels = {0: 'a', 1: 'a', 5: 'b', 6: 'b'}  # semi-supervised mode
n_clusters = 2  # Number of clusters

transform_type = 'std'  # preprocessing step
model_type = 'KMeans'  # clustering model

# Feature extraction
df_feats = feature_extraction(arr, batch_size=100, p=1)

# Feature selection
context = {'model_type': model_type, 'transform_type': transform_type}
top_feats = feature_selection(df_feats, labels=labels, context=context)
df_feats = df_feats[top_feats]

# Clustering
model = ClusterWrapper(n_clusters=n_clusters, model_type=model_type, transform_type=transform_type)
y_pred = model.fit_predict(df_feats)
print(y_pred.shape)

Working example

Demo: a script to apply time2feat on UEA & UCR multivariate time series dataset.

Dataset

All public multivariate time series datasets used in the paper can be downloaded from UEA & UCR Time Series Classification Repository. The demo code only support sktime formatted ts files ( see Cricket dataset).