mlr3oml

Package website: release | dev

OpenML integration to the mlr3 ecosystem.

What is mlr3oml?

OpenML is an open-source platform that facilitates the sharing and dissemination of machine learning research data. All entities on the platform have unique identifiers and standardized (meta)data that can be accessed via an open-access REST API or the web interface. mlr3oml allows to work with the REST API through R and integrates OpenML with the mlr3 ecosystem. Note that some upload options are currently not supported, use the OpenML package package for this.

As a brief demo, we show how to access an OpenML task, convert it to an mlr3::Task and associated mlr3::Resampling, and conduct a simple resample experiment.

library(mlr3oml)
library(mlr3)

# Download and print the OpenML task with ID 145953
oml_task = otsk(145953)
oml_task

## <OMLTask:145953>
##  * Type: Supervised Classification
##  * Data: kr-vs-kp (id: 3; dim: 3196x37)
##  * Target: class
##  * Estimation: crossvalidation (id: 1; repeats: 1, folds: 10)

# Access the OpenML data object on which the task is built
oml_task$data

## <OMLData:3:kr-vs-kp> (3196x37)
##  * Default target: class

# Convert the OpenML task to an mlr3 task and resampling
task = as_task(oml_task)
resampling = as_resampling(oml_task)

# Conduct a simple resample experiment
rr = resample(task, lrn("classif.rpart"), resampling)
rr$aggregate()

## classif.ce 
##  0.0319181

Besides working with objects with known IDs, data of interest can also be queried using listing functions. Below, we search for datasets with 10 - 20 features, 100 to 10000 observations and 2 classes.

odatasets = list_oml_data(
  number_features = c(10, 20),
  number_instances = c(100, 10000),
  number_classes = 2
)

odatasets[, c("data_id", "name")]

##      data_id
##   1:      13
##   2:      15
##   3:      29
##   4:      49
##   5:      50
##  ---        
## 238:   44767
## 239:   45039
## 240:   45063
## 241:   45562
## 242:   45568
##                                                                              name
##   1:                                                                breast-cancer
##   2:                                                                     breast-w
##   3:                                                              credit-approval
##   4:                                                                      heart-c
##   5:                                                                  tic-tac-toe
##  ---                                                                             
## 238: Click_prediction_small_seed_4_nrows_2000_nclasses_10_ncols_100_stratify_True
## 239:                                                             compas-two-years
## 240:                                                              credit-approval
## 241:                                                                seismic-bumps
## 242:                                                         telco-customer-churn

Feature Overview

• Datasets, tasks, flows, runs, and collections can be downloaded from OpenML and are represented as R6 classes.
• OpenML objects can be easily converted to the corresponding mlr3 counterpart.
• Filtering of OpenML objects can be achieved using listing functions.
• Downloaded objects can be cached by setting the mlr3oml.cache option.
• Both the arff and parquet filetype for datasets are supported.

Documentation

• Start by reading the Large-Scale Benchmarking chapter from the mlr3 book.
• The package website contains a getting started guide.
• The OpenML API documentation is also a good resource.

Bugs, Questions, Feedback

mlr3oml is a free and open source software project that encourages participation and feedback. If you have any issues, questions, suggestions or feedback, please do not hesitate to open an “issue” about it on the GitHub page!

In case of problems / bugs, it is often helpful if you provide a “minimum working example” that showcases the behaviour (but don’t worry about this if the bug is obvious).