load_from_csv.py

#!/usr/bin/env python3
"""
Kabyle-English Road Traffic Code Dataset - CSV Loader
Loads from dataset.csv and converts to HuggingFace format
"""

import pandas as pd
from datasets import load_dataset, Dataset, DatasetDict

def load_from_csv(csv_path="dataset.csv"):
    """Load dataset from CSV and convert to HuggingFace format"""
    df = pd.read_csv(csv_path)

    # Convert to list of dicts with proper structure
    records = []
    for _, row in df.iterrows():
        record = {
            "id": row["id"],
            "translation": {
                "en": row["english"],
                "kab": row["kabyle"]
            },
            "category": row["category"],
            "subcategory": row["subcategory"],
            "domain": "road_traffic",
            "complexity": row["complexity"],
            "tokens_en": int(row["tokens_en"]),
            "tokens_kab": int(row["tokens_kab"])
        }
        records.append(record)

    return records

def create_splits(csv_path="dataset.csv"):
    """Create train/validation/test splits from CSV"""
    records = load_from_csv(csv_path)

    # Split based on ID suffix (matching original stratified split)
    train_ids, val_ids, test_ids = [], [], []

    for r in records:
        idx = int(r["id"].split("_")[-1])
        cat = r["category"]

        # Reproduce stratified split logic
        if cat == "dangers":
            if idx % 10 < 7:
                train_ids.append(r)
            elif idx % 10 < 8:
                val_ids.append(r)
            else:
                test_ids.append(r)
        elif cat == "prohibitions":
            if idx % 10 < 6:
                train_ids.append(r)
            elif idx % 10 < 8:
                val_ids.append(r)
            else:
                test_ids.append(r)
        elif cat == "obligations":
            if idx % 10 < 5:
                train_ids.append(r)
            elif idx % 10 < 7:
                val_ids.append(r)
            else:
                test_ids.append(r)
        else:  # end_of_restrictions
            if idx % 10 < 4:
                train_ids.append(r)
            elif idx % 10 < 7:
                val_ids.append(r)
            else:
                test_ids.append(r)

    return {
        "train": Dataset.from_list(train_ids),
        "validation": Dataset.from_list(val_ids),
        "test": Dataset.from_list(test_ids)
    }

def load_dataset_from_csv(csv_path="dataset.csv"):
    """Main function - returns DatasetDict"""
    splits = create_splits(csv_path)
    return DatasetDict(splits)

if __name__ == "__main__":
    print("Chargement depuis dataset.csv...")
    dataset = load_dataset_from_csv()

    print(f"\nTrain: {len(dataset['train'])} examples")
    print(f"Validation: {len(dataset['validation'])} examples")
    print(f"Test: {len(dataset['test'])} examples")

    print("\n--- Exemple ---")
    ex = dataset['train'][0]
    print(f"ID: {ex['id']}")
    print(f"EN: {ex['translation']['en']}")
    print(f"KAB: {ex['translation']['kab']}")
    print(f"Category: {ex['category']}")