from datasets import load_dataset
import os
import jsonlines

# Load the dataset
dataset = load_dataset("ctu-aic/csfever_nli")

# Shuffle the test dataset and subsample 2/3 of the data
shuffled_test_dataset = dataset["test"].shuffle(seed=42)
subsampled_test_dataset = shuffled_test_dataset.select(range(int(len(shuffled_test_dataset) * 2 / 3)))

# Define the output directory and create it if it doesn't exist
output_dir = ".data/hf_dataset/csfever_nli"
os.makedirs(output_dir, exist_ok=True)

# Function to write the dataset split to a jsonl file
def write_jsonl(data, filepath):
    with jsonlines.open(filepath, mode='w') as writer:
        writer.write_all(data)

# Write the datasets to jsonl files
write_jsonl(dataset["train"], os.path.join(output_dir, "train.jsonl"))
write_jsonl(dataset["validation"], os.path.join(output_dir, "validation.jsonl"))
write_jsonl(subsampled_test_dataset, os.path.join(output_dir, "test.jsonl"))

print(f"Datasets saved to {output_dir}")