README.md

---
language:
- en
license: mit
size_categories:
- 10K<n<100K
pretty_name: siqa
tags:
- multiple-choice
- benchmark
- evaluation
configs:
- config_name: default
  data_files:
  - split: eval
    path: data/eval-*
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: id
    dtype: int32
  - name: question
    dtype: string
  - name: choices
    sequence: string
  - name: answerID
    dtype: int32
  splits:
  - name: eval
    num_bytes: 380631
    num_examples: 1954
  - name: train
    num_bytes: 6460849
    num_examples: 33410
  download_size: 3900341
  dataset_size: 6841480
---

# siqa Dataset

## Dataset Information
- **Original Hugging Face Dataset**: `lighteval/siqa`
- **Subset**: `default`
- **Evaluation Split**: `validation`
- **Training Split**: `train`
- **Task Type**: `multiple_choice`
- **Processing Function**: `process_siqa` 

## Processing Function
The following function was used to process the dataset from its original source:
```python
def process_siqa(example: Dict) -> Tuple[str, List[str], int]:
    """Process SocialIQA dataset example."""
    query = f"{example['context']} {example['question']}"
    
    # Get the original choices
    original_choices = [example['answerA'], example['answerB'], example['answerC']]
    correct_answer = original_choices[int(example["label"]) - 1]  # Convert 1-based index to 0-based
    
    # Find the new index of the correct answer after shuffling
    answer_index = original_choices.index(correct_answer)
    
    return query, original_choices, answer_index

```
## Overview
This repository contains the processed version of the siqa dataset. The dataset is formatted as a collection of multiple-choice questions.

## Dataset Structure
Each example in the dataset contains the following fields:
```json
{
  "id": 0,
  "question": "Tracy didn't go home that evening and resisted Riley's attacks. What does Tracy need to do before this?",
  "choices": [
    "make a new plan",
    "Go home and see Riley",
    "Find somewhere to go"
  ],
  "answerID": 2
}
```

## Fields Description
- `id`: Unique identifier for each example
- `question`: The question or prompt text
- `choices`: List of possible answers
- `answerID`: Index of the correct answer in the choices list (0-based)

## Loading the Dataset
You can load this dataset using the Hugging Face datasets library:
```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("DatologyAI/siqa")

# Access the data
for example in dataset['train']:
    print(example)
```

## Example Usage
```python
# Load the dataset
dataset = load_dataset("DatologyAI/siqa")

# Get a sample question
sample = dataset['train'][0]

# Print the question
print("Question:", sample['question'])
print("Choices:")
for idx, choice in enumerate(sample['choices']):
    print(f"{idx}. {choice}")
print("Correct Answer:", sample['choices'][sample['answerID']])
```