TVS Regression

Experiments with Time Varying Stochastic Regression

Installation

1. Create conda environment.

conda create -n tvsr python=3.8
conda activate tvsr

2. Install requirements

pip install -r requirements.txt

Example

An example of a univariate linear TVS regression can be found at: /notebooks/1_univariate_example.ipynb

TVS Regression Article

---

Abstract

Systems with stochastic time delay between the input and output present a number of unique challenges. Time domain noise leads to irregular alignments, obfuscates relationships and attenuates inferred coefficients. To handle these challenges, we introduce a maximum likelihood regression model that regards stochastic time delay as an 'error' in the time domain. For a certain subset of problems, by modelling both prediction and time errors it is possible to outperform traditional models. Through a simulated experiment of a univariate problem, we demonstrate results that significantly improve upon Ordinary Least Squares (OLS) regression.

The full article can be found at: /documentation/article/article.pdf