search

RJT1990 / pyflux

Open source time series library for Python
BSD 3-Clause "New" or "Revised" License
2.1k stars 240 forks source link

GASX fitting unstable #93

Open fccoelho opened 7 years ago

fccoelho commented 7 years ago

I am having issues fitting GASX (Poisson family) models with more than one exogenous variable.

MLE estimation yields a summary tabels full of NANs without any warning messages about what could be wrong. BBVI seems to go nowhere. Laplace fails complaining of NANs which do not exist on the dataset. MLE sometimes works for a single exogenous variable, but seems to have convergence issues yielding different estimates on different runs.

I am using the latest PyPI version.

Is it just my model, or is this behavior to be expected on GASX models.

RJT1990 commented 7 years ago

That's strange - could you post your code here and I'll investigate? It might be something to do with the Poisson family

fccoelho commented 7 years ago

The code for the model I am trying to fit is in this gist. The data is attached: data.csv.gz

fccoelho commented 7 years ago

@RJT1990, did you have a chance to run my example?