README.md

metadata

name: generalization-test
version: 1.0.0
description: Merge between 3 datasets.
configs:
  - config_name: clinc150
    default: true
    data_files:
      - split: train
        path: train_clinc150.csv
      - split: validation
        path: validation_clinc150.csv
      - split: test
        path: test_clinc150.csv
  - config_name: rostd+
    data_files:
      - split: train
        path: train_rostd+.csv
      - split: validation
        path: val_rostd+.csv
      - split: test
        path: test_rostd+.csv
license: openrail
task_categories:
  - text-classification
language:
  - en

This dataset merge 3 datasets and have two setup for experiments in generalisation for multi-class clasificacitino task.

• ID, near-OOD, covariate-shitf: CLINC150
• ID, near-OOD, covariate-shitf: ROSTD+OOD (fbreleasecoarse version)
• far-OOD Validation: SST2
• far-OOD Test: News Category (v3)