Release Notes 0.9
- orphan
Statsmodels is using now github to store the updated documentation which is available under http://www.statsmodels.org/stable for the last release, and http://www.statsmodels.org/devel/ for the development version.
Warning
API stability is not guaranteed for new features, although even in this case changes will be made in a backwards compatible way if possible. The stability of a new feature depends on how much time it was already in statsmodels master and how much usage it has already seen. If there are specific known problems or limitations, then they are mentioned in the docstrings.
The following major new features appear in this version.
Major refactoring of the underlying code. This has been in master for more than a year and is considered stable.
statsmodels.multivariate has gained several new statistical methods
Basic multivariate tests. ...
The implementation of MANOVA is based on a new _MultivariateOLS model which is currently an incomplete model class but contains the methods for multivariate hypothesis testing.
A new Factor class is available for
that's also new, ....
A new class for within, repeated measures ANOVA for balanced data AnovaRM is available in stats.anova.
New count models have been added as part of GSOC 2017. One additional set of models will be added for the next release.
The new models are:
- NegativeBinomialP (NBP): This is a generalization of NegativeBinomial that allows the variance power parameter to be specified in the range between 1 and 2. The current NegativeBinomial support NB1 and NB2 which are two special cases of NBP.
- GeneralizedPoisson (GPP): Similar to NBP this allows a large range of dispersion specification. GPP also allow some amount of under dispersion
- ZeroInflated ...
- A new class statsmodels.base.distributed_estimation.DistributedModel can help fit larger models by distributing the job.
- seasonal_decompose: works now vectorized for multiple time series and extrapolates trend to the beginning and end of the sample
- A minimal weighted least squares approach is used for the default IRLS fit which is much faster on larger data.
- var_weights and freq_weights are more robustly supported
- Significant clean-up/rework of the available families.
- Contributed by Kerby Shedden
- A new diagnostic tool using matplotlib
- Updated the implementation of Lilliefors' test in statsmodels.stats._lilliefors to include Lilliefors' test for exponential distributions
The VECM framework developed in this year's GSoC aims to add the support for non-stationary VAR processes to statsmodels. Currently, the following topics are implemented (see also this Jupyter-notebook):
- Parameter estimation,
- forecasting,
- testing for Granger-causality,
- testing for instantaneous causality (not on my initial todo-list and may be moved to vector_ar.py to extend the VAR * * * framework's functionality),
- impulse-response analysis, and
- lag order selection.
- see github issues
While most bugs are usability problems, there is now a new label type-bug-wrong for bugs that cause that silently incorrect numbers are returned. https://github.com/statsmodels/statsmodels/issues?q=label%3Atype-bug-wrong+is%3Aclosed
- In MixedLM, names for the random effects variance and covariance parameters have changed from, e.g. G RE to G Var or G x F Cov. This impacts summary output, and also may require modifications to user code that extracted these parameters from the fitted results object by name.
Besides receiving contributions for new and improved features and for bugfixes, important contributions to general maintenance came from
- Kevin Sheppard
- Tom Augsburger
and the general maintainer and code reviewer
- Josef Perktold
Additionally, many users contributed by participation in github issues and providing feedback.
Thanks to all of the contributors for the 0.8 release (based on git log):
Note
- Aleksandar Karakas
- Alex Fortin
- Alexander Belopolsky
- Brock Mendel
- Chad Fulton
- Christian Lorentzen
- Dave Willmer
- Evgeny Zhurko
- ftemme
- hugovk
- Jacob Kimmel
- Jamie Morton
- jarvmiller
- Jasmine Mou
- Joon Ro
- José Lopez
- Josef Perktold
- Joses W. Ho
- Kerby Shedden
- Kernc
- Kevin Sheppard
- kiwirob
- malickf
- Maxim Uvarov
- Michael Kaminsky
- Natasha
- Niels Wouda
- Pamphile ROY
- Peter Quackenbush
- Richard Höchenberger
- Roman Ring
- Scott Tsai
- Søren Fuglede Jørgensen
- Tom Augspurger
- Tony Jiang
- tvanzyl
- vegcev
- weizhongg
- Yichuan Liu
- zveryansky
These lists of names are automatically generated based on git log, and may not be complete.