Version 0.1.6 of PyFlux has been released on PyPi, including functionality for:
- GARCH models
- OLS/MLE/MAP results adjusted if user specifies transformed values.
- New plots for forecast (with fan chart option)
- GAS/ARMA/VAR code neatened and documented
- Adjustments to initialization conditions for GAS models
- There is now an option to plot the cumulative response for impulse responses in VAR models
I am possibly going to push another version online tomorrow with a few small updates (mainly concerning how results from estimation and prediction are stored). I will also update documentation and the notebooks as a few changes have been made to some input arguments.
There is an outstanding issue with VAR models at present where only OLS is reliable and fast enough. The other likelihood based options are both slow and often unreliable, as I haven’t implemented restrictions on the covariance matrix; I’ll look to fix this in near future.
In terms of the next steps (!) – I am currently working on Gaussian Process regression models. Hoping to get something ready next week; should be more exciting than the simple, classical stuff!