testthat::context("classifier")

test_that("can construct a pca classifier", {
  mat1 <- matrix(rnorm(10*15), 10, 15)
  pres <- pca(mat1)
  y <- rep(letters[1:4], length.out=10)
  cl <- classifier(pres, labels=y, new_data=mat1)
 
  p <- predict(cl, mat1)
  expect_true(ncol(p$prob) == 4)
  expect_true(nrow(p$prob) == 10)
})


test_that("can construct a pca classifier with colind", {
  mat1 <- matrix(rnorm(10*15), 10, 15)
  pres <- pca(mat1)
  y <- rep(letters[1:4], length.out=10)
  cl <- classifier(pres, labels=y, new_data=mat1[,1:5], colind=1:5)
  
  p <- predict(cl, mat1[,1:5])
  expect_true(ncol(p$prob) == 4)
  expect_true(nrow(p$prob) == 10)
  
  p2 <- predict(cl, mat1[1,1:5])
  expect_true(ncol(p2$prob) == 4)
  expect_true(nrow(p2$prob) == 1)
  
  p2 <- predict(cl, mat1[1,1:5], colind=1:5, metric="euclidean")
  expect_true(ncol(p2$prob) == 4)
  expect_true(nrow(p2$prob) == 1)
  
  p2 <- project(cl, mat1[1,1:5])
  expect_true(ncol(p2) == ncomp(pres))
  
})

test_that("can split a matrix", {
  mat1 <- matrix(rnorm(10*15), 10, 15)
  Y <- rep(letters[1:3], length.out=nrow(mat1))
  sm <- split_matrix(mat1,Y)
  expect_equal(3, length(sm))
})

test_that("can compute group means of rows of matrix", {
  mat1 <- matrix(rnorm(10*15), 10, 15)
  Y <- rep(letters[1:3], length.out=nrow(mat1))
  gm <- group_means(Y, mat1)
  expect_equal(nrow(gm), 3)
  
  Y <- letters[1:nrow(mat1)]
  gm <- group_means(Y, mat1)
  expect_equal(nrow(gm), length(Y))
})




test_that("feature_importance.classifier works with iris data and PCA", {
  # Data setup
  data(iris)
  X <- as.matrix(iris[, 1:4])
  labels <- iris[, 5]
  
  # Fit PCA
  pca_res <- pca(X, ncomp = 3)
  
  # Create classifier
  classifier <- classifier.projector(pca_res, labels = labels, new_data = X, knn = 1)
  
  # New data for prediction
  new_data <- X
  
  # Evaluate feature importance using the marginal approach
  importance_marginal <- feature_importance.classifier(classifier, new_data, ncomp = 2, 
                                                       true_labels = labels,
                                                       metric = "euclidean", fun = rank_score, 
                                                       approach = "marginal")
  print(importance_marginal)
  expect_true(all(importance_marginal$importance > 0))
  
  # Evaluate feature importance using the standalone approach
  importance_standalone <- feature_importance.classifier(classifier, new_data, ncomp = 2, 
                                                         true_labels = labels,
                                                         metric = "euclidean", fun = rank_score, 
                                                         approach = "standalone")
  print(importance_standalone)
  expect_true(all(importance_standalone$importance > 0))
})