sample_submission.py

"""
Sample submission script for Task 23: Property Inference.

Auxiliary dataset: https://huggingface.co/datasets/SprintML/Property_Inference
Download auxiliary.npz from there and place it in the same directory as this script.

Steps:
  1. Query the predict API for each model using auxiliary probe data
  2. Compute a confidence score per model (higher = more likely World A)
  3. Save predictions to submission.csv
  4. Submit submission.csv to the hackathon platform

Submission format (submission.csv):
    model_id,score
    0,0.73
    1,0.41
    ...

Where score is a float in [0, 1]:
    - Higher score = more likely World A (70% male training data)
    - Lower score  = more likely World B (50% male training data)
    - 0.5 = uncertain (this is the default for missing model_ids)

All 200 model_ids should be included. Missing ones default to 0.5.

API response format:
    {
        "model_id": 0,
        "batch_size": 100,
        "labels": [[4, 7], [2, 9], ...],          # top-2 predicted classes per record
        "probs":  [[0.52, 0.21], [0.44, 0.18], ...]  # top-2 softmax probabilities per record
    }

API rate limits:
    - Per model: 2 minute cooldown after a successful query
    - Failed requests: 2 minute cooldown
    - Max batch size: 500 records per request
"""

import csv
import json
import os

import numpy as np
import requests

# ── Configuration ──────────────────────────────────────────────────────────────
BASE_URL = "http://35.192.205.84:80"
API_KEY  = "YOUR_API_KEY_HERE"
TASK_ID  = "23-property-inference"

# Paths (relative to this script)
HERE          = os.path.dirname(os.path.abspath(__file__))
MODEL_IDS     = json.load(open(os.path.join(HERE, "model_ids.json")))
AUXILIARY_NPZ = np.load(os.path.join(HERE, "auxiliary.npz"))

PROBE_FEATURES = AUXILIARY_NPZ["features"].tolist()  # shape (10000, 10), already normalized
OUTPUT_CSV     = "submission.csv"

HEADERS = {"X-API-Key": API_KEY, "Content-Type": "application/json"}

# ── Query API ──────────────────────────────────────────────────────────────────
def query_model(model_id: int, features: list) -> dict:
    """
    Query the predict API for one model.

    Returns a dict with:
        labels: list[list[int]]   — top-2 predicted classes per record
        probs:  list[list[float]] — top-2 softmax probabilities per record
    """
    resp = requests.post(
        f"{BASE_URL}/23-property-inference/predict",
        headers=HEADERS,
        json={"model_id": model_id, "features": features},
        timeout=30,
    )
    resp.raise_for_status()
    data = resp.json()
    return {"labels": data["labels"], "probs": data["probs"]}


# ── Score computation (replace with your own method) ───────────────────────────
def compute_score(labels: list, probs: list) -> float:
    """
    Compute a confidence score in [0, 1] that a model belongs to World A.

    You have access to:
        labels: list[list[int]]   — top-2 predicted class indices per record
        probs:  list[list[float]] — top-2 softmax probabilities per record

    This baseline returns a random score — replace with your actual method.

    Returns:
        float in [0, 1] — higher means more likely World A
    """
    # Placeholder: random score. Replace with your actual method.
    score = float(np.random.uniform(0, 1))
    return float(np.clip(score, 0.001, 0.999))


# ── Main ───────────────────────────────────────────────────────────────────────
def main():
    predictions = {}

    # Use first 100 probe records (max batch size is 500)
    probe_batch = PROBE_FEATURES[:100]

    for model_id in MODEL_IDS:
        print(f"Querying model {model_id}...")
        try:
            result = query_model(model_id, probe_batch)
            score  = compute_score(result["labels"], result["probs"])
        except requests.exceptions.HTTPError as e:
            if e.response.status_code == 429:
                print(f"  Rate limited on model {model_id} — skipping (will default to 0.5)")
                continue
            else:
                raise
        predictions[model_id] = score
        print(f"  score={score:.4f}")

    # Write CSV
    with open(OUTPUT_CSV, "w", newline="") as f:
        writer = csv.writer(f)
        writer.writerow(["model_id", "score"])
        for model_id in MODEL_IDS:
            score = predictions.get(model_id, 0.5)
            writer.writerow([model_id, round(score, 6)])

    print(f"\nSaved {len(predictions)} predictions to {OUTPUT_CSV}")
    print(f"Missing models (defaulted to 0.5): {len(MODEL_IDS) - len(predictions)}")

    # Submit
    print("\nSubmitting...")
    with open(OUTPUT_CSV, "rb") as f:
        resp = requests.post(
            f"{BASE_URL}/submit/{TASK_ID}",
            headers={"X-API-Key": API_KEY},
            files={"file": (OUTPUT_CSV, f, "text/csv")},
            timeout=120,
        )
    print("Response:", resp.json())


if __name__ == "__main__":
    main()