leekgroup / postpi Goto Github PK

The package should contain these functions that the users see (the other helper functions can be hidden by not exporting them)

• iap_relate
  -inputs: test_dat, formula (with y as outcome, predictions as covariates))
  -outputs: rel_mod (the relationship model)

• iap_relate_plot

  • inputs: rel_mod
    -outputs: diagnostic plots (y and y_p for continuous along with model fit, for binary it could observed probability versus predicted)

• iap_glm

  • inputs: formula, valid_dat, rel_mod
  • outputs: regression model output from using the bootstrap approach (estimate, se, t-statistic, output in standard broom format)

• iap_lm_deriv

  • inputs: formula, valid_dat, rel_mod
  • outputs: regression model output from using iap_boot approach (estimate, se, t-statistic, output in standard broom format)

old

• iap_relate_plot

  • inputs: data set, prediction model, outcome variable, smoother choice (linear, loess)
  • outputs: plot of y versus predicted y for continuous data, plot of y versus predicted probability for binary

• iap_outcomes_plot

  • input: data set, prediction model, relationship model
    -output: plot of bootstrapped/simulated y versus real y

• iap_sim

  • input: number of covariates, number of samples (total), fraction in training/testing,validation, whether the outcome is binary or continous
  • output: data frame with columns outcome, covariate_1, covariate_2, etc., and group with values of either training, testing, or validation.

When building a prediction/relationship model the user should use caret: http://topepo.github.io/caret/index.html

When setting up the sampling scheme the user should use rsample: https://github.com/tidymodels/rsample

When setting up the inference model the user should use recipes: https://github.com/tidymodels/recipes

We should use yardstick to measure accuracy:
https://github.com/tidymodels/yardstick

Output for the models should be created in broom format:
https://github.com/tidymodels/broom

Hello! the link to the vignette is broken!

Can we use this guide please to make the package: https://github.com/jtleek/rpackages

We should include a step by step vignette on how to:

(1) Set up training/testing/validation
(2) Fit the model and use it.
(3) Diagnose when it is going wrong.
(4) Shows how to do this for many covariates.