compute_model_prediction: Create a model of a data set and compute predictions.

Description

Fit a 1d model, then compute predictions and (optionally) standard errors over an evenly spaced grid.

Usage

compute_model_prediction(
  x,
  formula,
  ...,
  model = NULL,
  se = FALSE,
  level = 0.95,
  n = 80L,
  domain = NULL,
  method
)
compute_smooth(x, formula, ..., span = 0.75, se = FALSE)

Value

A data frame with columns:

resp_: regularly spaced grid of n locations
pred_: predicted value from model
pred_lwr_ and pred_upr_: upper and lower bounds of confidence interval (if se = TRUE)
pred_se_: the standard error (width of the confidence interval) (if se = TRUE)

Arguments

x: Dataset-like object to model and predict. Built-in methods for data frames, grouped data frames and ggvis visualisations.
formula: Formula passed to modelling function. Can use any variables from data.
...: arguments passed on to model function
model: Model fitting function to use - it must support R's standard modelling interface, taking a formula and data frame as input, and returning predictions with predict. If not supplied, will use loess for <= 1000 points, otherwise it will use gam. Other modelling functions that will work include lm, glm and rlm.
se: include standard errors in output? Requires appropriate method of predict_grid, since the interface for returning predictions with standard errors is not consistent acrossing modelling frameworks.
level: the confidence level of the standard errors.
n: the number of grid points to use in the prediction
domain: If NULL (the default), the domain of the predicted values will be the same as the domain of the prediction variable in the data. It can also be a two-element numeric vector specifying the min and max.
method: Deprecated. Please use model instead.
span: Smoothing span used for loess model.

Details

compute_model_prediction fits a model to the data and makes predictions with it. compute_smooth is a special case of model predictions where the model is a smooth loess curve whose smoothness is controlled by the span parameter.

Examples

Run this code

# Use a small value of n for these examples
mtcars %>% compute_model_prediction(mpg ~ wt, n = 10)
mtcars %>% compute_model_prediction(mpg ~ wt, n = 10, se = TRUE)
mtcars %>% group_by(cyl) %>% compute_model_prediction(mpg ~ wt, n = 10)

# compute_smooth defaults to loess
mtcars %>% compute_smooth(mpg ~ wt)

# Override model to suppress message or change approach
mtcars %>% compute_model_prediction(mpg ~ wt, n = 10, model = "loess")
mtcars %>% compute_model_prediction(mpg ~ wt, n = 10, model = "lm")

# Set the domain manually
mtcars %>%
  compute_model_prediction(mpg ~ wt, n = 20, model = "lm", domain = c(0, 8))

# Plot the results
mtcars %>% compute_model_prediction(mpg ~ wt) %>%
  ggvis(~pred_, ~resp_) %>%
  layer_paths()
mtcars %>% ggvis() %>%
  compute_model_prediction(mpg ~ wt) %>%
  layer_paths(~pred_, ~resp_)