tests/testthat/test-linear.R

In sgd: Stochastic Gradient Descent for Scalable Estimation

context("Linear regression")

test_that("MSE converges for linear models", {

  skip_on_cran()

  # Dimensions
  N <- 1e4
  d <- 5

  # Generate data.
  set.seed(42)
  X <- matrix(rnorm(N*d), ncol=d)
  theta <- rep(5, d+1)
  eps <- rnorm(N)
  y <- cbind(1, X) %*% theta + eps
  dat <- data.frame(y=y, x=X)

  get.mse <- function(method, lr) {
    sgd.theta <- sgd(y ~ ., data=dat, model="lm",
                     sgd.control=list(
                       method=method,
                       lr=lr,
                       npasses=10,
                       pass=T))
    mean((sgd.theta$coefficients - theta)^2)
  }

  expect_true(get.mse("sgd", "one-dim") < 1e-2)
  expect_true(get.mse("sgd", "adagrad") < 1e-2)
  #expect_true(get.mse("sgd", "rmsprop") < 1e-2)
  expect_true(get.mse("implicit", "one-dim") < 1e-2)
  #expect_true(get.mse("implicit", "d-dim") < 1e-2)
  expect_true(get.mse("implicit", "adagrad") < 1e-2)
  #expect_true(get.mse("implicit", "rmsprop") < 1e-2)
  expect_true(get.mse("asgd", "one-dim") < 1e-2)
  expect_true(get.mse("asgd", "adagrad") < 1e-2)
  expect_true(get.mse("asgd", "rmsprop") < 1e-2)
  expect_true(get.mse("ai-sgd", "one-dim") < 1e-2)
  expect_true(get.mse("ai-sgd", "d-dim") < 1e-2)
  expect_true(get.mse("ai-sgd", "adagrad") < 1e-2)
  expect_true(get.mse("ai-sgd", "rmsprop") < 1e-2)
  expect_true(get.mse("momentum", "one-dim") < 1e-2)
  #expect_true(get.mse("momentum", "d-dim") < 1e-2)
  #expect_true(get.mse("momentum", "adagrad") < 1e-2)
  #expect_true(get.mse("momentum", "rmsprop") < 1e-2)
  expect_true(get.mse("nesterov", "one-dim") < 1e-2)
  #expect_true(get.mse("nesterov", "d-dim") < 1e-2)
  #expect_true(get.mse("nesterov", "adagrad") < 1e-2)
  #expect_true(get.mse("nesterov", "rmsprop") < 1e-2)
})