GSoC 2026: mlr3 Submission

1. Implementation Strategy

Testing the as_task() converter across two structurally different OpenML tasks to ensure robust metadata mapping and feature type preservation.

Task 3: kr-vs-kp

Binary classification validation
Categorical feature handling

Task 6: letter

Multiclass (26 labels) validation
Integer/Numeric feature mapping

2. Executed Code

library(mlr3)
library(mlr3oml)

# Fetching and converting tasks
task1 <- as_task(mlr3oml::otsk(3))
task2 <- as_task(mlr3oml::otsk(6))

print(task1)
print(task2)

3. Execution Logs & Object Output

The following logs confirm successful API communication and TaskClassif object creation.

Console Output: Task 1 (kr-vs-kp)

── <TaskClassif> (3196x37) ─────────────────
• Target: class
• Target classes: won (positive class, 52%), nowin (48%)
• Properties: twoclass
• Features (36):
  • fct (36): bkblk, bknwy, bkon8, bkona, bkspr, bkxbq, ...

Console Output: Task 2 (letter)

INFO  [23:36:50.250] Retrieving JSON {url: `openml.org/api/v1/json/data/6`}
INFO  [23:36:51.445] Retrieving ARFF {url: `openml.org/data/v1/download/6/letter.arff`}

── <TaskClassif> (20000x17) ────────────────
• Target: class
• Target classes: U (4%), D (4%), P (4%), T (4%), M (4%), ... + 16 more
• Properties: multiclass
• Features (16):
  • int (16): high, onpix, width, x.bar, x.box, x.ege, ...

4. Validation Summary

Results:
- Confirmed automatic target detection for binary and multiclass tasks.
- Verified fct (factor) mapping for categorical datasets.
- Verified memory efficiency for 20,000+ row datasets.