tests/testthat/test-TrainedInputs.R

In MachineShop: Machine Learning Models and Tools

## Trained Inputs

test_that("testing of trained inputs", {
  skip_if_not(TEST_TRAINING)
  with_parallel({

    library(recipes)

    df <- ICHomes
    fo1 <- sale_amount ~ sale_year + built + style + construction
    fo2 <- sale_amount ~ sale_year + base_size + bedrooms + basement

    rec1 <- recipe(sale_amount ~ ., data = ICHomes)
    rec2 <- rec1 %>%
      step_center(all_numeric_predictors()) %>%
      step_scale(all_numeric_predictors()) %>%
      step_pca(all_numeric_predictors(), id = "pca")

    sel_fo <- SelectedInput(fo1, fo2, data = df)
    sel_mat <- SelectedInput(
      model.matrix(~ sale_year + built + style + construction, ICHomes)[, -1],
      model.matrix(~ sale_year + base_size + bedrooms + basement, ICHomes)[, -1],
      y = ICHomes$sale_amount
    )
    sel_mf <- SelectedInput(
      ModelFrame(fo1, data = df),
      ModelFrame(fo2, data = df)
    )
    sel_mo_mf <- SelectedInput(
      ModelSpecification(fo1, data = df, model = GLMModel),
      ModelSpecification(fo2, data = df, model = GBMModel)
    )
    sel_rec <- SelectedInput(rec1, rec2)
    sel_mo_rec <- SelectedInput(
      ModelSpecification(rec1, model = GLMModel),
      ModelSpecification(rec2, model = GBMModel)
    )
    tun_rec <- TunedInput(rec2, grid = expand_steps(pca = list(num_comp = 1:3)))


    models <- c(
      "GLMModel" = GLMModel,
      "TunedModel(GLMNet)" = TunedModel(GLMNetModel),
      "SelectedModel(SVMModel, RangerModel)" = SelectedModel(SVMModel, RangerModel)
    )


    for (name in names(models)) {

      model <- models[[name]]

      ## model fitting of selected model frames
      expect_is(fit(sel_fo, model = model), "MLModelFit")
      expect_is(fit(sel_mat, model = model), "MLModelFit")
      expect_is(fit(sel_mf, model = model), "MLModelFit")

      ## model fitting of selected recipes
      expect_is(fit(sel_rec, model = model), "MLModelFit")

      ## model fitting of tuned recipe
      expect_is(fit(tun_rec, model = model), "MLModelFit")

    }


    ## model fitting of selected inputs
    model <- models[[1]]
    expect_is(fit(sel_mo_mf), "MLModelFit")
    expect_is(fit(sel_mo_rec), "MLModelFit")


    ## resampling of selected inputs
    model <- models[[1]]
    expect_is(resample(sel_mf, model = model), "Resample")
    expect_is(resample(sel_mo_mf), "Resample")
    expect_is(resample(sel_rec, model = model), "Resample")
    expect_is(resample(sel_mo_rec), "Resample")


    ## resampling of tuned inputs
    model <- models[[1]]
    expect_is(resample(tun_rec, model = model), "Resample")

  })
})