tests/testthat/test-classify_predict_matches.R

In niallbenj/footballstats: Classify and predict football results

context('test-classify_all.R')

# Reset DB
KEYS$RED$FLUSHDB()

test_that("No predictions to be made", {

  # If no data is found, then it will be assigned NULL
  noResults <- KEYS %>%
    footballstats::predict_matches(
      datModel = footballstats::nnModel,
      cMethod = "neuralnetwork",
      NULL
    )

  expect_equal( noResults, 0 )

})

test_that('Classify all - end to end from adding data to classifying and predicting (both methods).', {

  # Get data ready
  raw.dat <- footballstats::data.2017 %>%
    subset(footballstats::data.2017$comp_id %>% `==`("1204"))

  # NEED TO UPDATE DATE (otherwise handled in amatch_info)
  raw.dat$formatted_date %<>% format("%d.%m.%Y")

  train.dat <- raw.dat[1:359, ]
  test.dat <- raw.dat[360:370, ]
  com.metrics <- footballstats::total.metrics[1:(train.dat %>% nrow), ]

  # Add data to redis
  match.data <- KEYS %>% footballstats::amatch_info(train.dat)

  # Add commentaries
  com.metrics %>% footballstats::comm_from_metrics()

  # Order the match data
  match.data %<>% footballstats::order_matchdata()

  # Set up league information for predictions
  KEYS %>%
    footballstats::create_table(
      matchData = match.data
    )

  # Update weekly positions
  KEYS %>%
    footballstats::weekly_positions()

  # Create the predictions here
  KEYS$LOG_PRED <- TRUE

  # Set up grids
  GRIDS <- list(
    DAYS = 3,
    GRID_PTS = 8,
    GRID_BOUND = 0.2,
    DECAY = 5000,
    TOTAL_PERC = 1,
    NN_REP = 1,
    NN_THRESH = 0.005,
    XG_ROUNDS = 100,
    XG_DEPTH = 10,
    XG_ETA = 0.2,
    XG_GAMMA = 2
  )

  # Loop over the two different methods here
  for (i in 2:2) {
    next
    # Choose the right model
    datModel <- if (i == 1) {
      footballstats::nnModel
    } else {
      footballstats::total.metrics[1:300, ] %>%
        footballstats::optimize_variables(
          GRIDS = GRIDS,
          optimizeModels = FALSE,
          types = "xgboost"
        ) %>%
        `[[`("xgb") %>%
        `[[`("model")
    }

    KEYS %>% footballstats::predict_matches(
      datModel = datModel,
      cMethod = if (i == 1) "neuralnetwork" else "xgboost",
      test.dat
    )

    predictions <- "csdm_pred:1204:*" %>%
      KEYS$RED$KEYS() %>%
      purrr::flatten_chr() %>%
      strsplit(split = "[:]") %>%
      purrr::map(5) %>%
      purrr::flatten_chr() %>%
      as.integer %>%
      sort

    # Number of predictions match up?
    expect_equal( predictions %>% length, test.dat %>% nrow )

    # Make sure all the IDs match up
    matchIDs <- test.dat$id %>% as.integer %>% sort
    for (j in 1:(test.dat %>% nrow)) expect_equal( predictions[j], matchIDs[j] )
  }

  KEYS$LOG_PRED <- FALSE
  # Reset db
  KEYS$RED$FLUSHDB()

  expect_equal(1, 1)

})


test_that('Classify all - end to end from adding data to classifying and predicting (xgboost).', {
#
#   # Get data ready
#   raw.dat <- footballstats::data.2017 %>%
#     subset(footballstats::data.2017$comp_id %>% `==`("1204"))
#
#   # NEED TO UPDATE DATE (otherwise handled in amatch_info)
#   raw.dat$formatted_date %<>% format("%d.%m.%Y")
#
#   train.dat <- raw.dat[1:359, ]
#   test.dat <- raw.dat[360:370, ]
#   com.metrics <- footballstats::total.metrics[1:(train.dat %>% nrow), ]
#
#   # Add data to redis
#   match.data <- KEYS %>% footballstats::amatch_info(train.dat)
#
#   # Add commentaries
#   com.metrics %>% footballstats::comm_from_metrics()
#
#   # Order the match data
#   match.data %<>% footballstats::order_matchdata()
#
#   # Set up league information for predictions
#   KEYS %>%
#     footballstats::create_table(
#       matchData = match.data
#     )
#   KEYS %>% footballstats::weekly_positions()
#
#   # Prepare the xgboost model
#   GRIDS <- list(
#     DAYS = 3,
#     GRID_PTS = 8,
#     GRID_BOUND = 0.2,
#     DECAY = 5000,
#     TOTAL_PERC = 1,
#     NN_REP = 1,
#     NN_THRESH = 0.005,
#     XG_ROUNDS = 100,
#     XG_DEPTH = 10,
#     XG_ETA = 0.2,
#     XG_GAMMA = 2
#   )
#
#   # Build a small xgboost model..
#   res <- footballstats::total.metrics[1:300, ] %>%
#     footballstats::optimize_variables(
#       GRIDS = GRIDS,
#       optimizeModels = FALSE,
#       types = "xgboost"
#     )
#
#   # Create the predictions here
#   KEYS$LOG_PRED <- TRUE
#   KEYS %>% footballstats::predict_matches(
#     datModel = res$xgb$model,
#     cMethod = "xgboost",
#     test.dat
#   )
#   KEYS$LOG_PRED <- FALSE
#
#   predictions <- "csdm_pred:1204:*" %>%
#     KEYS$RED$KEYS() %>%
#     purrr::flatten_chr() %>%
#     strsplit(split = "[:]") %>%
#     purrr::map(5) %>%
#     purrr::flatten_chr() %>%
#     as.integer %>%
#     sort
#
#   # Number of predictions match up?
#   expect_equal( predictions %>% length, test.dat %>% nrow )
#
#   # Make sure all the IDs match up
#   matchIDs <- test.dat$id %>% as.integer %>% sort
#   for (i in 1:(test.dat %>% nrow)) {
#     expect_equal( predictions[i], matchIDs[i] )
#   }
#
#   # Reset db
#   KEYS$RED$FLUSHDB()

  expect_equal(1, 1)
})