OpenDataScience Europe 2021 talk - Patrick Schratz

code.R

## ----mlr3-config, echo = FALSE---------------------------------------------------------------------------
lgr::get_logger("bbotk")$set_threshold("warn")
lgr::get_logger("mlr3")$set_threshold("warn")


## ----example, fig.show="hide"----------------------------------------------------------------------------
library("mlr3verse", quietly = TRUE)
set.seed(42)

# example tasks
tasks <- tsks(c("iris", "german_credit"))
# from {mlr3learners}
learners <- lrns(c("classif.rpart",
  "classif.ranger"))

# run a cross-val
bmg <- benchmark_grid(
  tasks, learners,
  rsmp("cv")
)
bmr <- benchmark(bmg)

# visualize by classification error
autoplot(bmr, measure = msr("classif.ce"))


## ----mlr3spatial-ex-1------------------------------------------------------------------------------------
library("mlr3")
library("mlr3learners")
library("mlr3spatial")

tif <- system.file("tif/L7_ETMs.tif",
  package = "stars"
)
l7data <- stars::read_stars(tif)

# create mlr3 backend from sf data
backend <- as_data_backend(l7data)


## ----mlr3spatial-ex-2------------------------------------------------------------------------------------
# create a "Random Forest" learner and train it
learner <- lrn("regr.ranger")
task <- as_task_regr(backend, target = "layer.1")

rows_train <- sample(1:task$nrow, 1000)
rows_pred <- setdiff(1:task$nrow, rows_train)

learner$train(task, row_ids = rows_train)


## ----mlr3spatial-ex-3, warning=FALSE, results=FALSE------------------------------------------------------
# set the output file and predict with the learner
pred <- predict_spatial(task, learner, format = "stars")


## ----mlr3spatial-ex-31-----------------------------------------------------------------------------------
pred


## ----mlr3spatial-ex-4, out.width="50%", fig.align='center'-----------------------------------------------
plot(pred, col = c("#440154FF", "#443A83FF", "#31688EFF",
  "#21908CFF", "#35B779FF", "#8FD744FF", "#FDE725FF"))


## ----mlr3spatiotempcv-ex---------------------------------------------------------------------------------
library("mlr3spatiotempcv")

# create 'sf' object from example data
data_sf <- sf::st_as_sf(ecuador, coords = c("x", "y"), crs = 32717)


## ---- echo=FALSE, out.width="100%", out.height="75%"-----------------------------------------------------
mapview::mapview(data_sf)


## ----mlr3spatiotempcv-ex11-------------------------------------------------------------------------------
# create ClassifST task
task <- TaskClassifST$new("ecuador_sf", backend = data_sf,
  target = "slides", positive = "TRUE"
)
print(task)


## ----mlr3spatiotempcv-ex2--------------------------------------------------------------------------------
library("mlr3learners")
library("ranger")
task <- tsk("ecuador")

learner <- lrn("classif.ranger", predict_type = "prob")
resampling_sp <- rsmp("repeated_spcv_coords",
  folds = 4, repeats = 2
)

rr_sp <- resample(
  task = task,
  learner = learner,
  resampling = resampling_sp
)

rr_sp$aggregate(measures = msr("classif.ce"))


## ----mlr3spatiotempcv-ex3, fig.retina=1, dpi=150, fig.dim=c(8, 3)----------------------------------------
autoplot(resampling_sp, task, fold_id = c(1:2), size = 0.7) *
  ggplot2::scale_y_continuous(breaks = seq(-3.97, -4, -0.01)) *
  ggplot2::scale_x_continuous(breaks = seq(-79.06, -79.08, -0.01))