File size: 4,796 Bytes
fc6d86b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 | """
Sample submission script for Task 23: Property Inference.
Auxiliary dataset: https://huggingface.co/datasets/SprintML/Property_Inference
Download texas_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)
- Do not submit scores of exactly 0 or 1
All 140 model_ids should be included. Missing ones default to 0.5.
API rate limits:
- Per model: 15 minute cooldown after a successful query
- Failed requests: 2 minute cooldown
- Max batch size: 100 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, "texas_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) -> list:
"""Query the predict API and return predicted labels."""
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()
return resp.json()["labels"]
# ── Score computation (replace with your own method) ───────────────────────────
def compute_score(labels_a: list, labels_b: list = None) -> float:
"""
Compute a confidence score in [0, 1] that a model belongs to World A.
This baseline compares label distributions across two queries — you should
replace this with your own property inference 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)
probe_batch = PROBE_FEATURES[:100]
for model_id in MODEL_IDS:
print(f"Querying model {model_id}...")
try:
labels = query_model(model_id, probe_batch)
score = compute_score(labels)
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
# 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()
|