piecewise_cors: Piecewise Correlations
In brshallo/piececor: Tidy Piecewise Correlations

Description Usage Arguments Details Value Examples

View source: R/piecewise_cors.R

Independently for each variable of interest ... and .target, calculate the piecewise correlation. Observations included for each correlation calculation are segmented based on local extrema (minima / maxima) of the smoothed model fit.

piecewise_cors(
  data,
  .target,
  ...,
  custom_model_spec = NULL,
  fit_formula = ".target ~ s(...)",
  cor_function = "~cor(.x, method = 'spearman')[2,1]"
)

`data`	Dataframe containing columns of interest.
`.target`	Column name of target (unquoted). Numeric with sufficient unique observations.
`...`	Column names of variables of interest (unquoted), or a tidy selection specification (see: https://dplyr.tidyverse.org/reference/select.html). Separate models are created for each variable passed into `...`. Numeric with sufficient unique observations.
`custom_model_spec`	`NULL` or model specification from the `{parsnip}` package. Default is `NULL` in which case the model specification will be: parsnip::gen_additive_mod() %>% parsnip::set_engine("mgcv", method = "REML") %>% parsnip::set_mode("regression") (At this point) must be a generalized additive model with `{mgcv}` engine. See for details on parameters: https://parsnip.tidymodels.org/reference/details_gen_additive_mod_mgcv.html
`fit_formula`	Character string. Pseudo representation of formula passed to `parsnip::fit()` of model specification. Default is `".target ~ s(...)"`. Note that `.target` and `...` must be included in any custom `fit_formula`.
`cor_function`	Character string of a lambda function passed to `purrr::map*()` to calculate correlation between `.target` and variables of interest at piecewise segments of observations. Default is: `"~cor(.x, method = 'spearman')[2,1]"` Should be written in tidyverse friendly shortcut lambda notation (like above), as opposed to traditional lambda function notation: `"function(x) cor(x, method = 'spearman')[2,1]"` Must evaluate to a numeric vector of length one. Could pass in functions specification unrelated to correlation.

If the smoothing model has no local minima / maxima, correlation is calculated across all observations.

Named list of named list of models $mods and named list of dataframes $cors for each variable of interest specified in .... Each dataframe in .cors has the columns:

gtoe: (greater than or equal to) lt: (less than) data: list column containing dataframes of observations within segment n_obs: number of rows in data segment cor: value returned by cor_function for segment of observations