sigllm 0.0.2

Signals plus LLMs

An open source project from Data to AI Lab at MIT.

SigLLM

Using Large Language Models (LLMs) for time series anomaly detection.

• Homepage: https://github.com/sintel-dev/sigllm

Overview

SigLLM is an extension of the Orion library, built to detect anomalies in time series data using LLMs. We provide two types of pipelines for anomaly detection:

• Prompter: directly prompting LLMs to find anomalies in time series.
• Detector: using LLMs to forecast time series and finding anomalies through by comparing the real and forecasted signals.

For more details on our pipelines, please read our paper.

Quickstart

Install with pip

The easiest and recommended way to install SigLLM is using pip:

pip install sigllm

This will pull and install the latest stable release from PyPi.

In the following example we show how to use one of the SigLLM Pipelines.

Detect anomalies using a SigLLM pipeline

We will load a demo data located in tutorials/data.csv for this example:

import pandas as pd

data = pd.read_csv('data.csv')
data.head()

which should show a signal with timestamp and value.

     timestamp      value
0   1222840800   6.357008
1   1222862400  12.763547
2   1222884000  18.204697
3   1222905600  21.972602
4   1222927200  23.986643
5   1222948800  24.906765

In this example we use gpt_detector pipeline and set some hyperparameters. In this case, we set the thresholding strategy to dynamic. The hyperparameters are optional and can be removed.

In addtion, the SigLLM object takes in a decimal argument to determine how many digits from the float value include. Here, we don't want to keep any decimal values, so we set it to zero.

from sigllm import SigLLM

hyperparameters = {
    "orion.primitives.timeseries_anomalies.find_anomalies#1": {
        "fixed_threshold": False
    }
}

sigllm = SigLLM(
    pipeline='gpt_detector',
    decimal=0,
    hyperparameters=hyperparameters
)

Now that we have initialized the pipeline, we are ready to use it to detect anomalies:

anomalies = sigllm.detect(data)

:warning: Depending on the length of your timeseries, this might take time to run.

The output of the previous command will be a pandas.DataFrame containing a table of detected anomalies:

        start         end  severity
0  1225864800  1227139200  0.625879

Resources

Additional resources that might be of interest:

• Learn about Orion.
• Read our paper.

Citation

If you use SigLLM for your research, please consider citing the following paper:

Sarah Alnegheimish, Linh Nguyen, Laure Berti-Equille, Kalyan Veeramachaneni. Can Large Language Models be Anomaly Detectors for Time Series?.

@inproceedings{alnegheimish2024sigllm,
  title={Can Large Language Models be Anomaly Detectors for Time Series?},
  author={Alnegheimish, Sarah and Nguyen, Linh and Berti-Equille, Laure and Veeramachaneni, Kalyan},
  booktitle={2024 IEEE International Conferencze on Data Science and Advanced Analytics (IEEE DSAA)},
  organization={IEEE},
  year={2024}
}

History

0.0.2 - 2024-10-24

New Prompter pipeline.

• Test README with GPT – Issue #20 by @sarahmish
• Mistral-prompter – Issue #19 by @Linh-nk

0.0.1 - 2024-09-25

First sigllm release to PyPI: https://pypi.org/project/sigllm/

• Add README – Issue #17 by @sarahmish
• Create a SigLLM API – Issue #13 by @sarahmish
• Add a Quick Example – Issue #12 by @sarahmish
• Forecasting Pipeline – Issue #11 by @sarahmish
• Refactor Transformation Primitives – Issue #7 by @sarahmish
• Forecasting Module – Issue #2 by @sarahmish