Sfhmm

Highlight

Since v0.7.x, the motor stepping analyses are reimplemented with Rust instead of Cython. This update improves computation time, compile safety and support of the latest python versions.

What's Changed
* Reformat by hanjinliu in https://github.com/hanjinliu/sfHMM/pull/17
* Use Rust instead of Cython by hanjinliu in https://github.com/hanjinliu/sfHMM/pull/18


**Full Changelog**: https://github.com/hanjinliu/sfHMM/compare/v0.6.0...v0.7.0

New Features
---------------
- `matplotlib` based interactive viewer is working! Call `view_in_qt()` function and GUI will be launched. Same function is available in `sfHMM1`, `sfHMMn` and motor versions.
python
sf.run_all(False)
sf.view_in_qt()


Improvements
----------------
- HMM optimization algorithms are executed with normalized data. Now sfHMM is fully scalable (I've tested in the range of 10^-6 - 10^6 and there were no difference between scales).
- Read/write functions in `sfHMM.io` are moved to member functions. Just make an instance and call `read`:
python
sf = sfHMM1(psf=0.002, sg0=13, krange=(2, 7))
sf.read("path/to/somewhere")

- Path to the source file is tagged to `sfHMM` object in attribute `source` if it is read by function `read`. Therefore you don't have to write down the path to the source directory again in `save` function. Results will be saved in the same directory as name "XXX-sfHMMresult" by default.
python
sf = sfHMM().read("path/to/somewhere")
sf.run_all()
sf.save() that's it!

Improvements
---------------
- Better scalability in GMM clustering. It will not return ill results for data with small values (like ~1e-2).
- More user friendly warnings and error handlings.
- `sfHMMnMotor` had a wrong `krange` estimation function so that it used to take very long time to execute GMM clustering. It is now fixed.
- `Cython` is moved to relative import so that it is not needed to be installed beforehand.

Future Plans
-------------
- HMM optimization algorithms are not scalable yet. I'll fix them soon but for now input data should be rescaled if they are very small (like ~1e-3).

First Release.