generate_subdata.py

import json
import random
from pathlib import Path
from collections import defaultdict, Counter


def build_subsample_then_random_split(
    input_json,
    output_dir,
    samples_per_task=(1, 3),
    train_ratio=0.8,
    seed=42,
):
    random.seed(seed)

    input_json = Path(input_json)
    output_dir = Path(output_dir)
    output_dir.mkdir(parents=True, exist_ok=True)

    with open(input_json, "r") as f:
        data = json.load(f)

    # 1. Group by task
    task_to_samples = defaultdict(list)
    for item in data:
        task_to_samples[item["task"]].append(item)

    # 2. Subsample 1-3 samples per task
    sub_data = []
    task_sample_counts = {}

    for task, samples in task_to_samples.items():
        k = random.randint(samples_per_task[0], samples_per_task[1])
        k = min(k, len(samples))

        selected = random.sample(samples, k)
        sub_data.extend(selected)
        task_sample_counts[task] = k

    # 3. Random sample-level train/test split
    random.shuffle(sub_data)

    n_train = int(len(sub_data) * train_ratio)
    train_data = sub_data[:n_train]
    test_data = sub_data[n_train:]

    train_tasks = [item["task"] for item in train_data]
    test_tasks = [item["task"] for item in test_data]

    train_task_set = set(train_tasks)
    test_task_set = set(test_tasks)
    overlap_tasks = sorted(list(train_task_set & test_task_set))

    # 4. Save
    with open(output_dir / "train.json", "w") as f:
        json.dump(train_data, f, indent=2)

    with open(output_dir / "test.json", "w") as f:
        json.dump(test_data, f, indent=2)

    split_info = {
        "split_type": "subsample_per_task_then_sample_level_random_split",
        "seed": seed,
        "samples_per_task": list(samples_per_task),
        "train_ratio": train_ratio,

        "num_original_samples": len(data),
        "num_total_tasks": len(task_to_samples),
        "num_subsampled_samples": len(sub_data),

        "num_train_samples": len(train_data),
        "num_test_samples": len(test_data),

        "num_train_tasks": len(train_task_set),
        "num_test_tasks": len(test_task_set),
        "num_overlap_tasks": len(overlap_tasks),

        "overlap_tasks": overlap_tasks,
        "task_sample_counts_after_subsample": task_sample_counts,
        "train_task_counts": dict(Counter(train_tasks)),
        "test_task_counts": dict(Counter(test_tasks)),
    }

    with open(output_dir / "split_info.json", "w") as f:
        json.dump(split_info, f, indent=2)

    print(f"Original samples: {len(data)}")
    print(f"Total tasks: {len(task_to_samples)}")
    print(f"Subsampled samples: {len(sub_data)}")
    print(f"Train samples: {len(train_data)}")
    print(f"Test samples: {len(test_data)}")
    print(f"Train tasks: {len(train_task_set)}")
    print(f"Test tasks: {len(test_task_set)}")
    print(f"Overlap tasks: {len(overlap_tasks)}")
    print(f"Saved to: {output_dir}")


if __name__ == "__main__":
    build_subsample_then_random_split(
        input_json="./train_metadata.json",
        output_dir="./subfolder_exp_split",
        samples_per_task=(1, 3),
        train_ratio=0.8,
        seed=42,
    )