mlr3oml enables downloading OpenML datasets, tasks, flows, runs, and collections and allows to convert them to the corresponding mlr3 object. Furthermore, you can also obtain the data and the resampling for a given OpenML task. Caching can be enabled by setting the option "mlr3oml.cache". Uploading to OpenML is currently not supported, use the OpenML package package for this.
Short Demo
library("mlr3")
library("mlr3oml")
# be less verbose
lgr::get_logger("mlr3oml")$set_threshold("warn")
# retrieve data set as task from OML
tsk("oml", data_id = 31)## <TaskClassif:credit-g> (1000 x 21)
## * Target: class
## * Properties: twoclass
## * Features (20):
## - fct (13): checking_status, credit_history, employment,
## foreign_worker, housing, job, other_parties, other_payment_plans,
## own_telephone, personal_status, property_magnitude, purpose,
## savings_status
## - int (7): age, credit_amount, duration, existing_credits,
## installment_commitment, num_dependents, residence_since
# retrieve a regular task from OML
tsk("oml", task_id = 59)## <TaskClassif:iris> (150 x 5)
## * Target: class
## * Properties: multiclass
## * Features (4):
## - dbl (4): petallength, petalwidth, sepallength, sepalwidth
# retrieve resampling from OML
rsmp("oml", task_id = 59)## <ResamplingCustom>: Custom Splits
## * Iterations: 10
## * Instantiated: TRUE
## * Parameters: list()## [1] "iris"
odata$nrow## [1] 150
odata$ncol## [1] 5
head(odata$data)## sepallength sepalwidth petallength petalwidth class
## 1: 5.1 3.5 1.4 0.2 Iris-setosa
## 2: 4.9 3.0 1.4 0.2 Iris-setosa
## 3: 4.7 3.2 1.3 0.2 Iris-setosa
## 4: 4.6 3.1 1.5 0.2 Iris-setosa
## 5: 5.0 3.6 1.4 0.2 Iris-setosa
## 6: 5.4 3.9 1.7 0.4 Iris-setosa
backend = as_data_backend(odata)
backend## <DataBackendDataTable> (150x6)
## sepallength sepalwidth petallength petalwidth class ..row_id
## 5.1 3.5 1.4 0.2 Iris-setosa 1
## 4.9 3.0 1.4 0.2 Iris-setosa 2
## 4.7 3.2 1.3 0.2 Iris-setosa 3
## 4.6 3.1 1.5 0.2 Iris-setosa 4
## 5.0 3.6 1.4 0.2 Iris-setosa 5
## 5.4 3.9 1.7 0.4 Iris-setosa 6
## [...] (144 rows omitted)
# list oml data sets with 5 features and 50 - 200 instances
tab = list_oml_data(number_features = 5, number_instances = c(50, 200))
head(tab[, .(data_id, name)])## data_id name
## 1: 61 iris
## 2: 199 fruitfly
## 3: 214 baskball
## 4: 329 hayes-roth
## 5: 346 aids
## 6: 668 witmer_census_1980## <TaskClassif:credit-g> (1000 x 21)
## * Target: class
## * Properties: twoclass
## * Features (20):
## - fct (13): checking_status, credit_history, employment,
## foreign_worker, housing, job, other_parties, other_payment_plans,
## own_telephone, personal_status, property_magnitude, purpose,
## savings_status
## - int (7): age, credit_amount, duration, existing_credits,
## installment_commitment, num_dependents, residence_since
resampling = as_resampling(otask)
resampling## <ResamplingCustom>: Custom Splits
## * Iterations: 10
## * Instantiated: TRUE
## * Parameters: list()
otask$data## <OMLData:31:credit-g> (1000x21)
## * Default target: class
otask$name## [1] "Task 31: credit-g (Supervised Classification)"
otask$nrow## [1] 1000
otask$ncol## [1] 21
otask$task## NULL
otask$resampling## NULL
# list first 10 oml tasks
tab = list_oml_tasks(limit = 10)
tab[, .(task_id, data_id, name)]## task_id data_id name
## 1: 2 2 anneal
## 2: 3 3 kr-vs-kp
## 3: 4 4 labor
## 4: 5 5 arrhythmia
## 5: 6 6 letter
## 6: 7 7 audiology
## 7: 8 8 liver-disorders
## 8: 9 9 autos
## 9: 10 10 lymph
## 10: 11 11 balance-scale## [1] "Weka_3.7.5"
oflow$parameter## name data_type default_value
## 1: A flag
## 2: B flag
## 3: C option 0.25
## 4: J flag
## 5: L flag
## 6: M option 2
## 7: N option
## 8: O flag
## 9: Q option
## 10: R flag
## 11: S flag
## 12: U flag
# non-executable pseudo learner
learner = as_learner(oflow, "regr")
learner## <LearnerRegrOML100:oml.100>
## * Model: -
## * Parameters: list()
## * Packages: mlr3
## * Predict Types: [response]
## * Feature Types: -
## * Properties: -## <OMLData:952:prnn_fglass> (214x10)
## * Default target: type
orun$task## <OMLTask:3815>
## * Type: Supervised Classification
## * Data: prnn_fglass (id: 952; dim: 214x10)
## * Target: type
## * Estimation: crossvalidation (id: 1; repeats: 1, folds: 10)
orun$flow## <OMLFlow:3364>
## * Name: classif.boosting
## * Dependencies: mlr_2.8, adabag_4.1, rpart_4.1.10
head(orun$prediction)## repeat. fold row_id prediction truth
## 1: 0 0 33 window_float_glass window_float_glass
## 2: 0 0 52 window_float_glass window_float_glass
## 3: 0 0 67 window_float_glass window_float_glass
## 4: 0 0 49 window_float_glass window_float_glass
## 5: 0 0 41 window_float_glass window_float_glass
## 6: 0 0 5 window_non-float_glass window_float_glass
## confidence.containers confidence.tableware confidence.vehicle_glass
## 1: 0 0 0
## 2: 0 0 0
## 3: 0 0 0
## 4: 0 0 0
## 5: 0 0 0
## 6: 0 0 0
## confidence.vehicle_headlamp_glass confidence.window_float_glass
## 1: 0 1
## 2: 0 1
## 3: 0 1
## 4: 0 1
## 5: 0 1
## 6: 0 0
## confidence.window_non.float_glass
## 1: 0
## 2: 0
## 3: 0
## 4: 0
## 5: 0
## 6: 1
rr = as_resample_result(orun)
rr## <ResampleResult> of 10 iterations
## * Task: prnn_fglass
## * Learner: oml.3364
## * Warnings: 0 in 0 iterations
## * Errors: 0 in 0 iterations
# R6 class for collection
ocol = OMLCollection$new(232) # or
ocol = ocl(232)
ocol$tasks## id task data task_type target nrow ncol
## 1: 3 <OMLTask[25]> kr-vs-kp Supervised Classification class 3196 37
## 2: 6 <OMLTask[25]> letter Supervised Classification class 20000 17
## missing numeric symbolic binary task_splits
## 1: 0 0 37 35 crossvalidation
## 2: 0 16 1 0 crossvalidation