🎯 Easy Task: mlr3 OML Integration

👤 Contributor

Name: Aksh Kaushik

GitHub: @aksh08022006

Repository: github.com/aksh08022006/mlr3hf

Submission Date: February 5, 2026

🎨 Approach & Strategy

The easy task focuses on validating OpenML integration with mlr3 through the mlr3oml::otsk() function. The goal is to demonstrate that we can reliably download, parse, and convert OpenML datasets into structured mlr3::Task objects.

Why this approach?

• Canonical validation: The iris dataset (OpenML ID 59) is widely recognized and benchmarked, ensuring reproducibility across different environments.
• Edge case handling: A custom handcrafted dataset tests mixed column types, factor encoding, and role inference—critical for robust task construction.
• Minimal dependencies: Avoids bloated imports and ensures the solution is lightweight and maintainable.

📊 Test Datasets

💻 Implementation

The implementation demonstrates two complementary approaches to task creation:

Approach A: OpenML Integration via otsk()

Fetch and convert an OpenML dataset directly:

library(mlr3)
library(mlr3oml)

# Download iris dataset (OpenML ID 59) and convert to mlr3 Task
task_iris <- as_task(otsk(id = 59))
print(task_iris)

Approach B: Custom Task Construction

Demonstrate that as_task_classif() handles mixed data types and role assignment:

library(mlr3)

# Build small handcrafted dataset with mixed types
stationery_data <- data.frame(
  length_cm = c(14.5, 15.0, 13.8, 17.5, 18.2, 19.1),
  has_ink = factor(c("Yes", "Yes", "Yes", "No", "No", "No")),
  has_eraser = factor(c("No", "No", "No", "Yes", "Yes", "Yes")),
  body_material = factor(c("Plastic", "Plastic", "Metal", "Wood", "Wood", "Wood")),
  label = factor(c("Pen", "Pen", "Pencil", "Pencil", "Pen", "Pencil"))
)

# Convert to mlr3::Task with explicit target role
task_stationery <- as_task_classif(
  stationery_data, 
  target = "label", 
  id = "pen_vs_pencil"
)
print(task_stationery)

📈 Expected Output

When executed, both approaches produce structured mlr3::Task objects:

<TaskClassif: iris (150 x 5)>
  Target: Species
  Properties: multiclass
  Features (4): Sepal.Length, Sepal.Width, Petal.Length, Petal.Width

<TaskClassif: pen_vs_pencil (6x5)>
  Target: label
  Properties: multiclass
  Features (4): length_cm, has_ink, has_eraser, body_material

🔍 Why This Solution Works

This submission demonstrates end-to-end integration robustness across multiple scenarios:

1. OpenML Connectivity

By fetching datasets from OpenML, we prove that external data source integration is working. This validates the entire pipeline: HTTP requests, data parsing, and format conversion.

2. Type System Validation

The handcrafted dataset exercises the type coercion engine. as_task_classif() must correctly:

• Preserve numeric columns as numeric
• Recognize factors as categorical features
• Assign the target column as a classification outcome

3. Minimal, Focused Scope

Unlike heavy integration frameworks, this approach isolates and validates the core functionality: data ingestion + Task construction. No unnecessary dependencies or complex workflows.

🚀 How to Run Locally

Reproduce the entire workflow on your machine:

# 1. Install dependencies
Rscript -e "install.packages(c('mlr3','mlr3oml','mlr3data','testthat'), repos='https://cloud.r-project.org')"

# 2. Run the R code from the Implementation section above
# (Paste into an R console or script file)

# 3. Verify output matches the expected results above

Expected runtime: ~10–30 seconds (first run may download OpenML data)