Migrating from Previous Versions

If you were using a version of pymer4 older than v0.9.X you can refer to the notes and table below for how to migrate your workflow. These changes were motivated by two goals: (a) making a more consistent and intuitive API; (b) simplifying library maintenance and extensibility. In particular, we now offload all computations to R including multi-core parallelization (e.g. for bootstrapping) by leveraging various popular R libraries (e.g. parameters::model_parameters, lme4::bootMer). Python code is almost entirely a means to store and convert the inputs and outputs of a variety of R functions in the tidystats module using polars dataframes.

API Changes

New models classes	Old model classes
lm()	Lm()
glm()	Lm(); only family ='binomial'
lmer()	Lmer()
glmer()	Lmer(); with family kwarg

Function	New models API	Old model API
Create a model	model('y ~ x', data=polars_df)	Model('y ~ x', data=pandas_df)
Fit a model	.fit()	same
ANOVA table	.anova()	same
Marginal estimates & comparisons	.emmeans() / .empredict()	.post_hoc()
Parameter estimates	.params, .result_fit	.coef
Fit statistics	.result_fit_stats	.AIC, .BIC, etc
Fixed-effects LMMs/GLMMs	.fixef	.fixef
Random-effects LMMs/GLMMs	.ranef	.ranef
Random-effects-variances LMMs/GLMMs	.ranef_var	.ranef_var
Use various statistical functions	pymer4.tidystats (R-based)	pymer4.stats (python-based)

Deprecated Features

• Removed support for Lm2() models

• Removed plotting methods (please use seaborn or matplotlib)

• Removed permutation-based inference for, e.g. model.fit(permute=1000)

• Removed miscellaneous stats and simulation functions

Planned Improvements

You can see the updated development roadmap on github

• integration with scikit-learn Estimators and cross-validation tools

• additional models (e.g. ordinal regression via ordinal::clm)