HMC (version 1.1)

anchored_lasso_testing: Anchored test for two-sample mean comparison.

Description

Anchored test for two-sample mean comparison.

Usage

anchored_lasso_testing(
  sample_1,
  sample_2,
  pca_method = "sparse_pca",
  mean_method = "lasso",
  lasso_tuning_method = "min",
  num_latent_factor = 1,
  n_folds = 5,
  verbose = TRUE
)

Value

A list of test statistics.

test_statistics

Test statistics. Each entry corresponds to the test result of one principle component.

standard_error

Estimated standard error of test_statistics_before_studentization.

test_statistics_before_studentization

Similar to test_statistics but does not have variance = 1.

split_data

Intermediate quantities needed for further assessment and interpretation of the test results.

Arguments

sample_1

Group 1 sample. Each row is a subject and each column corresponds to a feature.

sample_2

Group 2 sample. Each row is a subject and each column corresponds to a feature.

pca_method

Methods used to estimate principle component The default is "sparse_pca", using sparse PCA from package PMA. Other choices are "dense_pca"---the regular PCA; and "hard"--- hard-thresholding PCA, which also induces sparsity.

mean_method

Methods used to estimate the discriminant direction. Default is logistic Lasso "lasso". Can also take value "lasso_no_truncation"

lasso_tuning_method

Method for Lasso penalty hyperparameter tuning. Default is "min", the minimizer of cross-validation error; users can also use "1se" for more sparse solutions.

num_latent_factor

The principle component that lasso coefficient anchors at. The default is PC1 = 1.

n_folds

Number of splits when performing cross-fitting. The default is 5, if computational time allows, you can try to set it to 10.

verbose

Print information to the console. Default is TRUE.

Examples

sample_size_1 <- sample_size_2 <- 300
true_mean_1 <- matrix(c(rep(1, 10), rep(0, 90)), ncol = 1)
true_mean_2 <- matrix(c(rep(1.5, 10), rep(0, 90)), ncol = 1)

sample_1 <- data.frame(MASS::mvrnorm(sample_size_1,
                               mu = true_mean_1,
                               Sigma = diag(1, 100)))
 sample_2 <- data.frame(MASS::mvrnorm(sample_size_2,
                               mu = true_mean_2,
                               Sigma = diag(1, 100)))
 result <- anchored_lasso_testing(sample_1, sample_2)
 result$test_statistics
 ##the test statistic. It should follow normal(0,1) when there is no difference between the groups.
 summarize_feature_name(result) 
 #summarize which features contribute to discriminant vectors (i.e. logistic lasso)
 extract_pc(result) # extract the estimated discriminant coefficients