Experiments with Time Varying Stochastic Regression
- Create
conda
environment.
conda create -n tvsr python=3.8
conda activate tvsr
- Install requirements
pip install -r requirements.txt
An example of a univariate linear TVS regression can be found at: /notebooks/1_univariate_example.ipynb
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