Discussing the article: "Neural Networks Made Easy (Part 85): Multivariate Time Series Forecasting"
Thank you for your hard work!
You have written quite a lot of articles already, your luggage of knowledge is growing faster than I have time to familiarise myself with its product.
Do you plan to write a review article in which you will briefly express your opinion about the methods you have described, share your experience of using this or that method?
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Check out the new article: Neural Networks Made Easy (Part 85): Multivariate Time Series Forecasting.
In this article, I would like to introduce you to a new complex timeseries forecasting method, which harmoniously combines the advantages of linear models and transformers.
The main idea of Client is move from attention over time to analyzing dependencies between variables and integrating a linear module into the model to better exploit variable dependencies and trend information, respectively.
The authors of the Client method creatively approached the solution to the timeseries forecasting problem. On the one hand, the proposed algorithm incorporates already familiar approaches. On the other hand, it rejects some well-established methods. The inclusion or exclusion of each individual block in the algorithm is accompanied by a series of tests. The tests demonstrate the feasibility of the decision taken from the point of view of the model effectiveness.
To solve the problem with the distribution bias, the authors of the method use reversible normalization with a symmetric structure (RevIN), which was discussed in the previous article. RevIN is first used to removes statistical information about the timeseries from the original data. After the model processes the data and generates forecast values, the statistical information of the original timeseries is restored in the forecast, which generally allows to increase the stability of model training and the quality of the forecast values of the timeseries.
Author: Dmitriy Gizlyk