FittedGridSearch: Fitted Models across a Tuning Grid of Hyper-parameters

In dmolitor/modelselection: Model Selection and Tuning Utilities

Fitted Models across a Tuning Grid of Hyper-parameters

Description

FittedGridSearch is an object containing fitted predictive models across a tuning grid of hyper-parameters returned by GridSearch$fit() as well as relevant model information such as the best performing model, best hyper-parameters, etc.

Public fields

best_idx

An integer specifying the index of $models that contains the best-performing model.

best_metric

The performance metric of the best model on the validation data.

best_model

The best performing predictive model.

best_params

A named list of the hyper-parameters that result in the optimal predictive model.

tune_params

Data.frame of the full hyper-parameter grid.

models

List of predictive models at every value of $tune_params.

metrics

Numeric list; Cross-validation performance metrics on each fold.

predictions

A list containing the predicted hold-out values on every fold.

Methods

Public methods

• FittedGridSearch$new()

• FittedGridSearch$clone()

---

Method new()

Create a new FittedGridSearch object.

Usage

FittedGridSearch$new(tune_params, models, metrics, predictions, optimize_score)

Arguments

tune_params

Data.frame of the full hyper-parameter grid.

models

List of predictive models at every value of $tune_params.

metrics

List of performance metrics on the validation data for every model in $models.

predictions

A list containing the predicted values on the validation data for every model in $models.

optimize_score

Either "max" or "min" indicating whether or not the specified performance metric was maximized or minimized to find the optimal predictive model.

Returns

An object of class FittedGridSearch.

---

Method clone()

The objects of this class are cloneable with this method.

Usage

FittedGridSearch$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

dmolitor/modelselection documentation built on Jan. 4, 2023, 1:08 p.m.