Tods Save

TODS: An Automated Time-series Outlier Detection System

Project README

TODS: Automated Time-series Outlier Detection System


Actions Status codecov


TODS is a full-stack automated machine learning system for outlier detection on multivariate time-series data. TODS provides exhaustive modules for building machine learning-based outlier detection systems, including: data processing, time series processing, feature analysis (extraction), detection algorithms, and reinforcement module. The functionalities provided via these modules include data preprocessing for general purposes, time series data smoothing/transformation, extracting features from time/frequency domains, various detection algorithms, and involving human expertise to calibrate the system. Three common outlier detection scenarios on time-series data can be performed: point-wise detection (time points as outliers), pattern-wise detection (subsequences as outliers), and system-wise detection (sets of time series as outliers), and a wide-range of corresponding algorithms are provided in TODS. This package is developed by DATA Lab @ Rice University.

TODS is featured for:

  • Full Stack Machine Learning System which supports exhaustive components from preprocessings, feature extraction, detection algorithms and also human-in-the loop interface.

  • Wide-range of Algorithms, including all of the point-wise detection algorithms supported by PyOD, state-of-the-art pattern-wise (collective) detection algorithms such as DeepLog, Telemanon, and also various ensemble algorithms for performing system-wise detection.

  • Automated Machine Learning aims to provide knowledge-free process that construct optimal pipeline based on the given data by automatically searching the best combination from all of the existing modules.

Examples and Tutorials


Cite this Work:

If you find this work useful, you may cite this work:

	title={TODS: An Automated Time Series Outlier Detection System}, 
	journal={Proceedings of the AAAI Conference on Artificial Intelligence}, 
	author={Lai, Kwei-Herng and Zha, Daochen and Wang, Guanchu and Xu, Junjie and Zhao, Yue and Kumar, Devesh and Chen, Yile and Zumkhawaka, Purav and Wan, Minyang and Martinez, Diego and Hu, Xia}, 
	year={2021}, month={May}, 


This package works with Python 3.7+ and pip 19+. You need to have the following packages installed on the system (for Debian/Ubuntu):

sudo apt-get install libssl-dev libcurl4-openssl-dev libyaml-dev build-essential libopenblas-dev libcap-dev ffmpeg

Clone the repository (if you are in China and Github is slow, you can use the mirror in Gitee):

git clone

Install locally with pip:

cd tods
pip install -e .


Examples are available in /examples. For basic usage, you can evaluate a pipeline on a given datasets. Here, we provide example to load our default pipeline and evaluate it on a subset of yahoo dataset.

import pandas as pd

from tods import schemas as schemas_utils
from tods import generate_dataset, evaluate_pipeline

table_path = 'datasets/anomaly/raw_data/yahoo_sub_5.csv'
target_index = 6 # what column is the target
metric = 'F1_MACRO' # F1 on both label 0 and 1

# Read data and generate dataset
df = pd.read_csv(table_path)
dataset = generate_dataset(df, target_index)

# Load the default pipeline
pipeline = schemas_utils.load_default_pipeline()

# Run the pipeline
pipeline_result = evaluate_pipeline(dataset, pipeline, metric)

We also provide AutoML support to help you automatically find a good pipeline for your data.

import pandas as pd

from axolotl.backend.simple import SimpleRunner

from tods import generate_dataset, generate_problem
from tods.searcher import BruteForceSearch

# Some information
table_path = 'datasets/yahoo_sub_5.csv'
target_index = 6 # what column is the target
time_limit = 30 # How many seconds you wanna search
metric = 'F1_MACRO' # F1 on both label 0 and 1

# Read data and generate dataset and problem
df = pd.read_csv(table_path)
dataset = generate_dataset(df, target_index=target_index)
problem_description = generate_problem(dataset, metric)

# Start backend
backend = SimpleRunner(random_seed=0)

# Start search algorithm
search = BruteForceSearch(problem_description=problem_description,

# Find the best pipeline
best_runtime, best_pipeline_result = search.search_fit(input_data=[dataset], time_limit=time_limit)
best_pipeline = best_runtime.pipeline
best_output = best_pipeline_result.output

# Evaluate the best pipeline
best_scores = search.evaluate(best_pipeline).scores


We gratefully acknowledge the Data Driven Discovery of Models (D3M) program of the Defense Advanced Research Projects Agency (DARPA)

Open Source Agenda is not affiliated with "Tods" Project. README Source: datamllab/tods
Open Issues
Last Commit
1 week ago

Open Source Agenda Badge

Open Source Agenda Rating