Conversation

JSON Baumgartner

@jasonbaumgartne

Can anyone recommend a good Python library that does anomaly detection for time-series data?

11:08 PM · Jun 25, 2019Twitter Web Client

Replying to

@jasonbaumgartne

what are you trying to do? you generally have to do these things from scratch...

1

1

JSON Baumgartner

@jasonbaumgartne

Replying to

I want to feed in documents where words / terms are tracked over time windows and set up alerts for when there is a spike (like if the mean for that window is over 4 standard deviations). I was wondering if there was such a beast already created for this.

1

3

Show replies

Replying to

@jasonbaumgartne

You can theoretically hack it with Prophet. towardsdatascience.com/anomaly-detect

1

18

JSON Baumgartner

@jasonbaumgartne

Replying to

Thanks! I'll check that out.

1

Show replies

Replying to

@jasonbaumgartne

Try using matrixprofile-ts on github. It can be very good at detecting anomalies

matrixprofile-ts/README.md at master · target/matrixprofile-ts

A Python library for detecting patterns and anomalies in massive datasets using the Matrix Profile - matrixprofile-ts/README.md at master · target/matrixprofile-ts

2

@igor@mastodon.social

Replying to

@jasonbaumgartne

Check out github.com/twitterdev/Gni for a good anomalous terms over time type anomaly detection thing

GitHub - twitterdev/Gnip-Trend-Detection: Trend detection algorithms for Twitter time series data

Trend detection algorithms for Twitter time series data - GitHub - twitterdev/Gnip-Trend-Detection: Trend detection algorithms for Twitter time series data

1

5

JSON Baumgartner

@jasonbaumgartne

Replying to

Thank you Igor!

1

𝚛𝚘𝚋𝚒𝚗_𝚎𝚍𝚠𝚊𝚛𝚍𝚜

Replying to

@jasonbaumgartne

Maybe find a python implementation of the Twitter breakout method anomaly.io/anomaly-detect

Matthew Alhonte

Replying to

@jasonbaumgartne

github.com/TDAmeritrade/s