get_results.py

import gradio as gr
import random
import firebase_admin
from firebase_admin import credentials
from firebase_admin import firestore
import uuid
import json
import os
from dotenv import load_dotenv
import re
import pandas as pd
from tqdm import tqdm

load_dotenv()

video_pairs = pd.read_csv('file_pairs.csv')[['file_name', 'vista_id', 'gem_id']].values.tolist()

my_credentials = {
    "type": "service_account",
    "project_id": "human-eval-c4f83",
    "private_key_id": os.environ.get("PRIVATE_KEY_ID"),
    "private_key": os.environ.get("PRIVATE_KEY").replace(r'\n', '\n'),
    "client_email": os.environ.get("CLIENT_EMAIL"),
    "client_id": os.environ.get("CLIENT_ID"),
    "auth_uri": "https://accounts.google.com/o/oauth2/auth",
    "token_uri": "https://oauth2.googleapis.com/token",
    "auth_provider_x509_cert_url": os.environ.get("AUTH_PROVIDER_X509_CERT_URL"),
    "client_x509_cert_url": os.environ.get("CLIENT_X509_CERT_URL")
}

if not firebase_admin._apps:
    cred = credentials.Certificate(my_credentials)
    firebase_admin.initialize_app(cred)

db = firestore.client()
# get all votes
votes = db.collection('votes').get()

gem_vote_count = {}
vista_vote_count = {}
no_preference_count = {}

qids = [
    'q1',
    'q2',
]

for vote in tqdm(votes):
    vote_dict = vote.to_dict()
    for qid in qids:
        category = qid
        if category not in gem_vote_count:
            gem_vote_count[category] = 0
            vista_vote_count[category] = 0
            no_preference_count[category] = 0
        
        if vote_dict[qid] == 'Video 1' and vote_dict['video1_id'] == vote_dict['gem_id']:
            gem_vote_count[category] += 1
        elif vote_dict[qid] == 'Video 1' and vote_dict['video1_id'] == vote_dict['vista_id']:
            vista_vote_count[category] += 1
        elif vote_dict[qid] == 'Video 2' and vote_dict['video2_id'] == vote_dict['gem_id']:
            gem_vote_count[category] += 1
        elif vote_dict[qid] == 'Video 2' and vote_dict['video2_id'] == vote_dict['vista_id']:
            vista_vote_count[category] += 1
        elif vote_dict[qid] == 'No preference':
            no_preference_count[category] += 1
        else:
            raise ValueError('Invalid vote: ' + str(vote_dict))

# Create a table with the results. One row for GEM, one row for Vista, one row for no preference

data = []
for qid in qids:
    category = qid
    data.append({
        'category': category,
        'gem': gem_vote_count[category],
        'vista': vista_vote_count[category],
        'no_preference': no_preference_count[category]
    })

df = pd.DataFrame(data)
df.to_csv('results.csv', index=False)