sandbox/test-kerasR.R

In tlverse/sl3: Pipelines for Machine Learning and Super Learning

context("test-kerasR.R -- time series methods")

if (FALSE) {
  setwd("..")
  setwd("..")
  getwd()
  library("devtools")
  document()
  load_all("./") # load all R files in /R and datasets in /data. Ignores NAMESPACE:
  # devtools::check() # runs full check
  setwd("..")
  install("sl3", build_vignettes = FALSE, dependencies = FALSE) # INSTALL W/ devtools:
}

library(reticulate)
library(sl3)
library(testthat)
library(origami)

set.seed(1)
attach(list(lag = stats::lag), name = "stats_lag_test_kludge", warn.conflicts = FALSE)
data(bsds)
bsds <- bsds[1:50, ]

data <- as.data.table(bsds)
data[, time := .I]

outcome <- "cnt"

folds <- origami::make_folds(data,
  fold_fun = folds_rolling_window, window_size = 20,
  validation_size = 15, gap = 0, batch = 10
)

node_list <- list(outcome = outcome, time = "time")

task <- sl3_Task$new(data, nodes = node_list, folds = folds)

# See which environments reticulate can see:
# reticulate:::conda_list()
# use_condaenv("r-reticulate")
# import("scipy")
# import("tensorflow")

# Install keras:
# install.packages("keras")
# keras::install_keras()

test_that("Lrnr_bilstm does what we expect", {
  have_foo <- reticulate::py_module_available("keras.models")

  if (!(have_foo)) {
    skip("keras.models not available for testing")
  } else {
    reticulate::import("keras.models")

    bilstm_learner <- Lrnr_bilstm$new(epochs = 1)
    bilstm_fit <- bilstm_learner$train(task)
    bilstm_preds <- bilstm_fit$predict(task)

    # At epochs=1 (saves time) the prediction is too variable to be tested
    # expect_true(sum(bilstm_preds)-118.5766 < 10^(-1))
    expect_equal(length(bilstm_preds), length(task$Y))
  }
})