codelionx / periodicity-detection

Detect dominant periodicity in equidistant time series

Hi there, I'm Sebastian Schmidl 👋

I'm a software engineer and PhD student at the chair for Information Systems (💻) at Hasso Plattner Institut for Digital Engineering (HPI). Currently, I'm working in the distributed computing research group, where we investigate computationally complex problems and how they can be solved in distributed environments.

🔭 Research Interests

• Scalable and reactive systems, especially using actor programming concepts
• Time series anomaly detection
• Distributed computing
• Data profiling

💻 Current open-source activity

• Maintainer @ TimeEval - An Evaluation Tool for Anomaly Detection Algorithms on Time Series
• Core developer @ aeon - A toolkit for conducting machine learning tasks with time series data

📫 How to reach me

• ✉️ by Mail (info [AT] sebastianschmidl [DOT] de)
• 🏠 Homepage
• 🏢 HPI homepage
• LinkedIn

---

periodicity-detection's Issues

Reduce dependencies and introduce soft-dependencies

For better compatibility with other modules, we should reduce the number of core dependencies. E.g., the dependency to matplotlib can be represented as a soft-dependency. This means that we do not require it to be installed to execute the code and import matplotlib just when the plotting code is executed.

Further candidates for soft-dependencies:

• spectrum is just required for findfrequency
• statsmodels is just required for find_length and autoperiod

pandas is not used at all in the code. Also check if we can get rid of the typing_extensions-dependency.

---

This leaves us with the following core dependencies:

numpy>=1.21.6
scipy>=1.7.3

Improve performance with Numba

We can improve the runtime of some methods by compiling the Python code with Numba's @jit. This should bring a significant speedup, especially for larger time series.