SelectionSimulator:

context("Simulation")
set.seed(7122018)

cov_mat_5 <- rbind( c(1.00, -0.80,  0.75, 0.00, 0.00),
                    c(-0.80,  1.00, -0.25, 0.00, 0.00),
                    c( 0.75, -0.25,  1.00, 0.00, 0.00),
                    c( 0.00,  0.00,  0.00, 1.00, 0.25),
                    c( 0.00,  0.00,  0.00, 0.25, 1.00))

coefficients_1 <- c(V1 = 1, V2 = 2, V3 = 3, V4 = 4, V5 = -5)

predict_gen <- NormalPredictorsGenerator$new(num_observations = 100,
                                             num_predictors = 5,
                                             norm_rand_var_sd = 2,
                                             covariance_matrix = cov_mat_5,
                                             predictor_names = names(coefficients_1))


response_calc_1 <- LinearNormalResponseCalculator$new(norm_rand_var_sd = 1,
                                                      coefficients = coefficients_1,
                                                      intercept = 9)

sim_data_gen <- SimulationDataGenerator$new(predict_gen, response_calc_1)

test_obj_1 <- Simulation$new(sim_data_gen, c(Step = StepLmWrapper$new, Step2 = Step2$new, Step3 = Step3$new), 5, 1)

test_obj_1$simulate()


# Changes if Step3 algorithm changes.
# Checks that Step3 is including V1V2V3 together at least once
test_that("triples_in_inclusion_order", {
  inclusion_orders_Step3 <- test_obj_1$get_inclusion_orders()[,"Step3"]
  triple_inclusions <- sapply(inclusion_orders_Step3, function(x) "V1V2V3" %in% x)
  expect_true(any(triple_inclusions))
})


test_that("fitted_coefficients", {
  X <- apply(test_obj_1$get_fitted_coefficients(), MARGIN = c(2,3), mean)
  expect_equal(mean(X["(Intercept)",]), 9, tolerance = 0.5)
  expect_equal(mean(X["V2",]), 2, tolerance = 0.5)
  expect_equal(mean(X["V3",]), 3, tolerance = 0.5)
  expect_equal(mean(X["V4",]), 4, tolerance = 0.5)
  expect_equal(mean(X["V5",]), -5, tolerance = 0.5)
})

max_cor <- test_obj_1$get_max_sample_correlation()[1:5,1:5]
min_cor <- test_obj_1$get_min_sample_correlation()[1:5,1:5]

test_that("sample_correlation_matrices_between_neg_1_and_1", {
  expect_true(all(max_cor <= 1))
  expect_true(all(min_cor <= 1))
  expect_true(all(max_cor >= -1))
  expect_true(all(min_cor >= -1))
})

test_that("sample_correlation_matrices_diagonals_are_one", {
  expect_true(all(diag(max_cor) == 1))
  expect_true(all(diag(min_cor) == 1))
})

test_sse <- test_obj_1$get_test_sse()
train_sse <- test_obj_1$get_training_sse()

test_that("test_train_sse_dimensions", {
  expect_equal(dim(test_sse), c(5,3))
  expect_equal(dim(train_sse), c(5,3))
})

test_that("test_train_no_NA_values", {
  expect_true(! any(is.na(test_sse)))
  expect_true(! any(is.na(train_sse)))
})