Package 'autogam'

Automate the creation of a Generalized Additive Model (GAM)

Description

autogam() is a wrapper for 'mgcv::gam()' that makes it easier to create high-performing Generalized Additive Models (GAMs). By entering just a dataset and the name of the outcome column as inputs, autogam() tries to automate the procedure of configuring a highly accurate GAM which performs at reasonably high speed, even for large datasets.

Usage

autogam(data, y_col, ..., bs = "cr")

Arguments

data	dataframe. All the variables in data will be used to predict y_col. To exclude any variables, assign as data only the subset of variables desired.
y_col	character(1). Name of the y outcome variable.
...	Arguments passed on to mgcv::gam().
bs	character(1). The default basis function for GAM smooths. See ?mgcv::smooth.terms for details. Whereas the default bs in mgcv is 'tp', autogam's default is 'cr', which is much faster and comparably accurate.

Value

Returns an mgcv::gam object, the result of predicting y_col from all other variables in data.

Examples

autogam(mtcars, 'mpg')

---

Generic autogam methods passed on to mgcv::gam methods

Description

An autogam object contains a gam element that is simply an mgcv::gam object. So, it supports all mgcv::gam methods by, in most cases, simply passing the gam element on to their corresponding mgcv::gam methods. Only the following methods have special specifications for autogam (see their dedicated documentation files for details): print.autogam().

Usage

## S3 method for class 'autogam'
anova(object, ...)

## S3 method for class 'autogam'
coef(object, ...)

## S3 method for class 'autogam'
cooks.distance(model, ...)

## S3 method for class 'autogam'
formula(x, ...)

## S3 method for class 'autogam'
influence(model, ...)

## S3 method for class 'autogam'
logLik(object, ...)

## S3 method for class 'autogam'
model.matrix(object, ...)

## S3 method for class 'autogam'
predict(object, ...)

## S3 method for class 'autogam'
residuals(object, ...)

## S3 method for class 'autogam'
vcov(object, ...)

Arguments

object, model	An object of class autogam.
...	other arguments
x	formula

Value

Returns the return object of the corresponding mgcv::gam method.

---

Plot Method for autogam Objects

Description

This function plots an autogam object. It calls the mgcv::gam object mgcv::plot.gam() method.

Usage

## S3 method for class 'autogam'
plot(x, ...)

Arguments

x	An object of class autogam.
...	Additional arguments passed to other methods.

Value

Same return object as mgcv::print.gam().

---

Print Method for autogam Objects

Description

This function prints an autogam object. It calls the mgcv::gam object print() method and then adds basic performance metrics from the autogam object:

• For models that predict numeric outcomes, it prints "MAE", the mean absolute error, and "Std. accuracy", the standardized accuracy (staccuracy) of the winsorized MAE relative to the mean absolute deviation.

• For models that predict binary outcomes, it prints "AUC", the area under the ROC curve.

Usage

## S3 method for class 'autogam'
print(x, ...)

Arguments

x	An object of class autogam.
...	Additional arguments passed to other methods.

Value

Invisibly returns the input object x.

---

Create a character string for a mgcv::gam formula

Description

Create a character string that wraps appropriate variables in a dataframe with s() smooth functions. Based on the datatype of each variable, it determines whether it is a numeric variable to be smoothed:

• Non-numeric: no smoothing.

• Numeric: determine knots based on the number of unique values for that variable:

  • ⁠<= 4⁠: no smoothing

  • ⁠5 to 19⁠ (inclusive): smooth function with knots equal to the floored half of the number of unique values. E.g., 6 unique values receive 3 knots, 7 will receive 3 knots, and 8 will receive 4 knots.

  • ⁠>= 20⁠: smooth function with no specified number of knots, allowing the gam() function to detect the appropriate number.

Usage

smooth_formula_string(
  data,
  y_col,
  smooth_fun = "s",
  bs = "cr",
  expand_parametric = TRUE
)

Arguments

data	dataframe. All the variables in data except y_col will be listed in the resulting formula string. To exclude any variables, assign as data only the subset of variables desired.
y_col	character(1). Name of the y outcome variable.
smooth_fun	character(1). Function to use for smooth wraps; default is 's' for the s() function.
bs	See documentation for autogam()
expand_parametric	logical(1). If TRUE (default), explicitly list each non-smooth (parametric) term. If FALSE, use . to lump together all non-smooth terms.

Value

Returns a single character string that represents a formula with y_col on the left and all other variables in data on the right, each formatted with an appropriate s() function when applicable.

Examples

smooth_formula_string(mtcars, 'mpg')

---

Summary Method for autogam Objects

Description

This function returns a summary of an autogam object. It calls the mgcv::gam object mgcv::summary.gam() method.

Usage

## S3 method for class 'autogam'
summary(object, ...)

Arguments

object	An object of class autogam.
...	Additional arguments passed to other methods.

Value

Same return object as mgcv::summary.gam().

---