Hello Everyone,
I am Anurag Bhat, a final year student at IIT Jodhpur. I am an open source enthusiast and usually contribute to Py-data repositories. I successfully completed GSoC’23 with SymPy where I worked on Expanding the functionalities of SymPy’s Control Module. Implementing a symbolic StateSpace model was one of my major contributions. You can have a look at my work here .
I would like to participate in GSoC’24 and express my interest towards the project - Implement New StateSpace Models. I have an understanding of StateSpace models in the contiguous time domain and have gone through the expected outcomes.
If mentors can share the current status of the project (through Github links of PRs) and resources to develop prerequisite understanding for the project it would be a great head start for me. Once I have some clarity, I should get started with making initial contributions.
Thanks and Regards.
Hi Anurag,
Thanks for your interest in the statespace project. The source is on the pymc-experimental repo here. There is an issue tracker with known bugs and desired extensions here. My suggested reading on the topic generally is Durbin and Koopsman 2012, Time Series Analysis by State Space Methods, chapters 2 and 3. This is a popular repository about Kalman filtering and linear state space models in Python, which is excellent background as well.
Other readings for specific aspects of the project:
- for exponential smoothing models check Forecasting: Principals and Practice chapter 8 , 8.5 in particular.
- Dynamic factor models: Durbin and Koopsman Chapter 3.7
- Multivariate structural models: Durbin and Koopsman Chapter 3.3
Edit: The statsmodels statespace docs are also helpful, because we followed their implementation quite closely. The developer of that package wrote a nice paper summarizing his approach.
Let me know if you have specific questions about anything related to the project/ssms generally.