sintel-dev/sigllm

LLMs for sintel

sigllm's Issues

Refactor Transformation Primitives

Transformation primitives

• Float2Scalar
  • turn floats into scalar integers.
  • this includes rescaling and taking n decimal points.
• Scalar2Float
  • turns scalars into floats.
  • reverse operation of Float2Scalar
• format_as_string
  • convert windows of integers to windows of string
• format_as_integer
  • convert windows of strings into windows of integer

Forecasting Module

Description

In addition to directly finding anomalies in time series data, forecasting can be utilized for the same task. In many approaches for unsupervised time series anomaly detection, forecasting has been used to generate an alternative signal which then gets compared to the original one. If there are discrepancies between the two signals, this is an indication of where the anomaly might be.

Expectation

Use large language models for time series forecasting.

prompt -> time series values
model -> any huggingface model
output -> forecasted time series