Refactor code to make it easier to implement non standard approaches:
[x] Subclasses for singleR (currently only one for standard implemented frequentist count modelling later add class for bayesian and mixed effects)
[x] Change methods to adjust for these subclasses
[x] Make it so that family type objects fully handle data (i.e. with zero one truncated or chao/zelterman)
[ ] Implement ratio regression
Further diagnostic information:
[ ] Null deviance
[x] deviance residuals (for models that do require a paper with theory)
[ ] ...
Fitting:
[ ] Bias Reduction compatibility with brglm2
[ ] Robust zero-truncated Poisson #5
[ ] Robust zero-truncated Geometric
[ ] EM algorithm
[ ] Estimation by full likelihood i.e. not observed likelihood
[x] Link functions in family functions
[ ] Additional measures for intercept only models to make them faster
Quality of life improvements:
[x] Prior weights as population counts
[ ] Class for multiple models
[ ] Multithreading (or possibly julia backend if it will be easier) in bootstrap and multiplication of $\boldsymbol{W}_{k}$ matrixes with pseudo-residuals and simulation of information information matrixes in negative binomials
[x] Update the simulate.singleR method and create simulate.singleRfamily, update.singleR methods.
singleRcaptureversion 0.3.0Refactor code to make it easier to implement non standard approaches:
singleR(currently only one for standard implemented frequentist count modelling later add class for bayesian and mixed effects)Further diagnostic information:
Fitting:
brglm2Quality of life improvements:
juliabackend if it will be easier) in bootstrap and multiplication of $\boldsymbol{W}_{k}$ matrixes with pseudo-residuals and simulation of information information matrixes in negative binomialssimulate.singleRmethod and createsimulate.singleRfamily,update.singleRmethods.Compatibility:
singleRclassModels:
Moved to singleRcaptureExtra package:
vglmzerotrunc