11annah / compint Goto Github PK

Home Page: https://11annah.github.io/CompInt/

R 1.68% Python 0.10% HTML 0.05% JavaScript 97.09% CSS 1.08%

compint's Introduction

CompInt

The goal of CompInt is to (eventually) allow users to compute all quantities from “A formal framework for generalized reporting methods in parametric settings” for a variety of model objects. Currently, this package is still in mid-development.

Installation

You can install the development version of CompInt from GitHub with:

# install.packages("pak")
pak::pak("11annah/CompInt")

# Alternatively, via devtools:
# install.packages("devtools")
devtools::install_github("11annah/CompInt")

Example

This is a basic example which shows you how to solve a common problem:

library(CompInt)

data(Affairs, package = "AER")
poisson.model <- glm(affairs ~ ., family = "poisson", data = Affairs)

draws_gME_religiousness <- get_gME(poisson.model, integration = assumption3(all_empirical()), reg_of_interest = "religiousness", seed = 100)
mean(draws_gME_religiousness)

draws_IE_children <- get_IE(poisson.model, integration = assumption3(all_empirical()), reg_of_interest = "children", seed = 100)
rowMeans(draws_IE_children)

compint's People

Contributors

Stargazers

Watchers

Forkers

m-muecke

compint's Issues

Complete documentation for all functions in "ModelStructures.R"

Special focus on occurences of #TOFIX

Output of the function binary_regs does not have the right format yet (I get errors)

make_g_theta: add options for non-GL(M)Ms

Make produce_model_definition cool

the seed works as expected on the same machine, but not on others...

I suspect it's because of torchquad...

Complete documentation for all functions in "g_theta.R"

Special focus on all instances of #TOFIX

Code for when the RI's reference category should be one that is not specified in the model

if("refcat" %in% ellipsisvars) in get_gME

Maybe make sure that the interaction notation stored in [["model_specification"]][["regs"]][["interactions"]][["notation"]] is not present in any regressor names?

Make a `various_to_compint` function!

Not only separate interactions, but any interactions between categorical variables need to be fixed

I'm not 100% sure where the issue is yet

Python warnings need to be handled better

Code for that should go into handle_warnings.R

Does the reference category just get added when there is no intercept?!

This must be found out

Think of a good assumptionstop stop message

See utils.R

Separate interactions : correct handling of the case with metric and categorical regs...

Specifically, should the metric interaction elements be stored as mRI_x_ ???

Adding non-empirical integration

Include case of separate_interactions=TRUE in "make_dummy_coded_data"

In utils.R

We need to decide on the format of "subset" and edit newdata_subset_merge accordingly

Specifically, in newdata_subset_merge (stored in utils.R) we may need to edit the subsetting process and DEFINETELY add a stop message for when subset has the wrong format

We need some "naming check" function

Specifically for

weird special characters
reg names that are fully contained in other reg names

Fix all instances of #TOFIX in the documentation

`coef`: For separating categorical interactions currently coef names are being used...

g_theta (make_linear_predictor): RI name is not sufficiently distinctive

When creating the linear predictor of the g_theta function, I am currently renaming the regressor of interest into "RI".
Unfortunately, this is an issue if either

RI or
the name of the regressor of interest (such as age)

appears within the name of another regressor...

The function is currently stored in in utils.R

We need to find a way to deal with regressors whose variable names have spaces.

Otherwise we don't get the regressor variable names from intersect(names(term_attr$dataClasses),term_attr$term.labels)

One option would be to forbid this encoding, but it did happen quite a lot in the Multi100 data...