This project generates data streams switching from different models that are based in dataset with different characteristics.
Download the source code from this repository and install the requirements from requirements.txt using PyPi.
The Python code uses the R rugarch library (version 1.4.1) through the rpy2 wrapper.
A path to a functioning R environment with the above-mentioned version of rugarch installed needs to be given in the config file config.yaml.
To generate a synthetic set, run python -m src.generator . Datasets to represent market data with different states, and changes between these states need to be specified beforehand in config.yaml. Parameter ranges to explore to fit models to these states, and the type of shifts, are also specified in this config file.
For more information about this generator, see the PhD thesis Adaptive Algorithms For Classification On High-Frequency Data Streams: Application To Finance.