tests/testthat/test_exp_bayes_ab.R

In exploratory-io/exploratory_func: R package for Exploratory

context("test exp_bayes_ab")
test_that("test exp_bayes_ab test with summary output", {
  set.seed(0)
  data_a <- data.frame(
    access_count = round(runif(50) * 100 + 100)
  ) %>%
    dplyr::mutate(click = 0.2 + 0.8 * runif(nrow(.)), `gro up` = "a")

  data_a$click[[5]] <- NA # inject NA for test

  data_b <- data.frame(
    access_count = round(runif(20) * 20 + 20)
  ) %>%
    dplyr::mutate(click = 0.3 + 0.7 * runif(nrow(.)), `gro up` = "b")

  full_data <- dplyr::bind_rows(data_a, data_b) %>%
    mutate(grouping = rbinom(n(), 1, 0.3)) %>%
    group_by(grouping)
  full_data <- full_data %>%
    dplyr::mutate(converted_true=as.integer(access_count*click), converted_false=as.integer(access_count*(1-click))) %>%
    tidyr::gather(key="conver ted", value="count col", converted_true, converted_false) %>%
    dplyr::mutate(`conver ted` = `conver ted`=="converted_true")
  full_data_logical_ab <- full_data %>% dplyr::mutate(`gro up` = `gro up` == "a")

  # revert A/B - character case
  model_df <- exp_bayes_ab(full_data, `conver ted`, `gro up`, `count col`, prior_mean=0.1, prior_sd=0.01)
  res <- model_df %>% tidy_rowwise(model, type = "summary", pretty.name = TRUE)
  res <- model_df %>% tidy_rowwise(model, type = "improvement")
  res <- model_df %>% tidy_rowwise(model, type = "posteriors")
  res <- model_df %>% tidy_rowwise(model, type = "prior")
  res <- model_df %>% tidy_rowwise(model, type = "observed")
  expect_equal(colnames(res),c("grouping","ab_identifier","converted","observed"))
})

test_that("test exp_bayes_ab test with summary output", {
  set.seed(0)
  data_a <- data.frame(
    access_count = round(runif(50) * 100 + 100)
  ) %>%
    dplyr::mutate(click = 0.2 + 0.8 * runif(nrow(.)), `gro up` = "a")

  data_a$click[[5]] <- NA # inject NA for test

  data_b <- data.frame(
    access_count = round(runif(20) * 20 + 20)
  ) %>%
    dplyr::mutate(click = 0.3 + 0.7 * runif(nrow(.)), `gro up` = "b")

  full_data <- dplyr::bind_rows(data_a, data_b) %>%
    mutate(grouping = rbinom(n(), 1, 0.3)) %>%
    group_by(grouping)
  full_data <- full_data %>%
    dplyr::mutate(converted_true=as.integer(access_count*click), converted_false=as.integer(access_count*(1-click))) %>%
    tidyr::gather(key="conver ted", value="count col", converted_true, converted_false) %>%
    dplyr::mutate(`conver ted` = `conver ted`=="converted_true")
  full_data_logical_ab <- full_data %>% dplyr::mutate(`gro up` = `gro up` == "a")

  # revert A/B - character case
  ret1 <- exp_bayes_ab(full_data, `conver ted`, `gro up`, `count col`, prior_mean=0.1, prior_sd=0.01, type = "summary")
  ret2 <- exp_bayes_ab(full_data, `conver ted`, `gro up`, `count col`, prior_mean=0.1, prior_sd=0.01, revert_ab = T, type = "summary")
  expect_equal(ret1$ab_identifier, c("a", "b", "a", "b"))
  expect_equal(ret2$ab_identifier, c("b", "a", "b", "a"))

  # revert A/B - logical case
  ret1 <- exp_bayes_ab(full_data_logical_ab, `conver ted`, `gro up`, `count col`, prior_mean=0.1, prior_sd=0.01, type = "summary")
  ret2 <- exp_bayes_ab(full_data_logical_ab, `conver ted`, `gro up`, `count col`, prior_mean=0.1, prior_sd=0.01, revert_ab = T, type = "summary")
  expect_equal(ret1$ab_identifier, c(TRUE,FALSE,TRUE,FALSE))
  expect_equal(ret2$ab_identifier, c("FALSE","TRUE","FALSE","TRUE"))

  #full_data <- full_data %>% rename(`access count`=access_count, `cli ck`=click)
  ret <- exp_bayes_ab(full_data, `conver ted`, `gro up`, `count col`, prior_mean=0.1, prior_sd=0.01, type = "summary")
  expect_equal(nrow(ret), 4)

  # without prior
  ret <- exp_bayes_ab(full_data, `conver ted`, `gro up`, `count col`, type = "summary")
  expect_equal(nrow(ret), 4)

  # without count with NULL
  ret <- exp_bayes_ab(full_data, `conver ted`, `gro up`, NULL, type = "summary")
  expect_equal(nrow(ret), 4)

  # without count
  ret <- exp_bayes_ab(full_data, `conver ted`, `gro up`, , type = "summary")
  expect_equal(nrow(ret), 4)
})

test_that("test exp_bayes_ab_aggregated", {
  df <- tibble::tibble(cat=c('A','B'), n=c(100,200), cr=c(0.22, 0.2))
  # revert A/B - character case
  model_df <- df %>% exp_bayes_ab_aggregated(cat, cr, n, prior_mean=0.1, prior_sd=0.01, seed=0)
  res <- model_df %>% tidy_rowwise(model, type = "summary", pretty.name = TRUE)
  expect_equal(res$`Chance of Being Better`, c(0.330, 0.670), tolerance=0.005)
  res <- model_df %>% tidy_rowwise(model, type = "improvement")
  res <- model_df %>% tidy_rowwise(model, type = "posteriors")
  res <- model_df %>% tidy_rowwise(model, type = "prior")
  res <- model_df %>% tidy_rowwise(model, type = "observed")
  expect_equal(colnames(res),c("ab_identifier","converted","observed"))
})

test_that("test exp_bayes_ab_aggregated with multiple rows per group", {
  # Here we test the case where a group (A or B) has more than 1 row.
  # It's not the most common use we expect, but this should also work.
  df <- tibble::tibble(cat=rep(c('A','B'),2), n=rep(c(100,200),2), cr=rep(c(0.22, 0.2),2))
  # revert A/B - character case
  model_df <- df %>% exp_bayes_ab_aggregated(cat, cr, n, prior_mean=0.1, prior_sd=0.01, seed=0)
  res <- model_df %>% tidy_rowwise(model, type = "summary", pretty.name = TRUE)
  expect_equal(res$`Chance of Being Better`, c(0.254, 0.746), tolerance=0.005)
  expect_equal(res$`Rows`,c(200,400)) # Number of rows should be added up.
  expect_equal(res$`Conversion Rate`,c(0.22,0.2)) # Ratios should be weight-averaged.
  res <- model_df %>% tidy_rowwise(model, type = "improvement")
  res <- model_df %>% tidy_rowwise(model, type = "posteriors")
  res <- model_df %>% tidy_rowwise(model, type = "prior")
  res <- model_df %>% tidy_rowwise(model, type = "observed")
  expect_equal(colnames(res),c("ab_identifier","converted","observed"))
})

test_that("test exp_bayes_ab_aggregated with more than 2 groups", {
  df <- tibble::tibble(cat=c('A','B','C'), n=c(100,200,300), cr=c(0.22, 0.2, 0.2))
  expect_error({
    model_df <- df %>% exp_bayes_ab_aggregated(cat, cr, n, prior_mean=0.1, prior_sd=0.01, seed=0)
  }, "A/B must be 2 unique identifiers")
})

test_that("test exp_bayes_ab_aggregated with only group A", {
  df <- tibble::tibble(cat=c('A'), n=c(100), cr=c(0.22))
  expect_error({
    model_df <- df %>% exp_bayes_ab_aggregated(cat, cr, n, prior_mean=0.1, prior_sd=0.01, seed=0)
  }, "A/B must be 2 unique identifiers")
})

#TODO: do the same set of tests as test_do_bayes_ab.R