In alexpghayes/reimagine: Scratch Work and Notes for a Reimagined Interface to Linear Models

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

library(magrittr)

mle_fit <- new_linear() %>%
  fit(mtcars, NULL)

lasso_fit <- new_linear(coefs = "lasso") %>%
  fit(mtcars, NULL)

mle_fit
lasso_fit

coef(mle_fit)
coef(lasso_fit)

class(mle_fit)
class(lasso_fit)

OLD NOTES

reimagine

Currently I'm just keeping some rough notes here on what I think linear modelling could look like in R.

things that people may want to estimate

• coefficients
• uncertainty in coefficients
• residual error scale

predictions

• confidence intervals for mean
• prediction intervals for new data points

estimators people may want to use

• OLS
• LASSO
• RIDGE
• horseshoe priors!
• HC1, HC2, etc etc

diagnostics

• convergence: dependent on estimator
• statistical

goodness of fit / predictive performance measures

• RMSE/MAE/etc etc for various resampling strategies
• R Squared
• Adj R Squared
• AIC / BIC
• F-statistic for model

convenience / utility functions

• write_model: produce latex for the model that was fit
• general form of the model
• specific model with coefficients from fit

interface

alexpghayes/reimagine documentation built on May 9, 2019, 7:31 p.m.