lin_least_squares_train_test: lin_least_squares_train_test

In rguin26/HW4:

Description

Performs linear regression using ordinary least squares (OLS) or weighted least squares (WLS), with either an intercept included or not, for only a portion of the total number of instances provided. The remaining instances are set aside for testing. This works with both simple and multiple linear regression. It calculates the beta coefficients of the least squares linear regression equation that is generated with the training set, and then it passes the testing set instances to generate predictions of that data which was previously unknown during the model training stage.

Usage

lin_least_squares_train_test(x, y, intercept = TRUE, weighted = FALSE, train_set_prop = 0.8)

Arguments

x	matrix, dataframe, or vector of all predictor/predictors and its/their respective values
y	vector of target values from the matrix of known predictor values
intercept	TRUE by default, it computes the beta coefficients with an intercept included; if it is set to FALSE, then no intercept is used
weighted	FALSE by default, it computes the beta coefficients using ordinary least squares (OLS); if it is set to TRUE, then the beta coefficients are calculated using weighted least squares (WLS)
train_set_prop	set to 0.8 by default, it randomly selects this proportion of instances from x and y, of identical indexes, to use for training the linear least squares model, and then it uses the remaining instances of x and y for testing the model

Details

works only with numeric data

Value

A list of objects, including calculations of the estimates for the beta coefficients for each predictor, along with residuals and model evaluation metrics for both the training and testing sets

• beta - vector of beta coefficients generated using the training subset of the initial data with values corresponding to their respective column in x, and starting with (intercept) if intercept = TRUE

• training_fitted_values - vector consisting of the training fitted values of the training subset of x used to build the model, calculated simply as the model's prediction value for each instance in the training subset

• training_residuals - vector consisting of the residuals of each instance in the training subset of x, calculated as the training fitted value minus the actual value for each instance in the training subset of x

• training_model_eval_metrics - vector consisting of the training subset's evaluation metrics of the model, including sum of squared errors (SSE), mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), r-squared, and adjusted r-squared

• testing_fitted_values - vector consisting of the testing fitted values of the testing subset of x, calculated simply as the model's prediction value for each instance in the testing subset

• testing_residuals - vector consisting of the residuals of each instance in the testing subset of x, calculated as the testing fitted value minus the actual value for each instance in the testing subset of x

• testing_model_eval_metrics - vector consisting of the testing subset's evaluation metrics of the model, including sum of squared errors (SSE), mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), r-squared, and adjusted r-squared

Examples

# Example 1
x <- data.frame(matrix(sample(100000, 200*5, replace=TRUE), ncol = 5))
y <- sample(100, 200, replace=TRUE)
model_stats <- lin_least_squares_train_test(x, y)
model_stats$beta
model_stats$training_model_eval_metrics
model_stats$testing_model_eval_metrics

rguin26/HW4 documentation built on Dec. 22, 2021, 3:09 p.m.