app.py

import os
import json

import datetime
from apscheduler.schedulers.background import BackgroundScheduler
import gradio as gr
import pandas as pd
from huggingface_hub import HfApi, snapshot_download

from utils import make_clickable_model
from utils import make_clickable_user

DATASET_REPO_URL = "https://huggingface.co/datasets/pkalkman/drlc-leaderboard-data"
DATASET_REPO_ID = "pkalkman/drlc-leaderboard-data"
HF_TOKEN = os.environ.get("HF_TOKEN")

block = gr.Blocks()
api = HfApi(token=HF_TOKEN)

# Read the environments from the JSON file
with open('envs.json', 'r') as f:
    rl_envs = json.load(f)


def download_leaderboard_dataset():
    # Download the dataset from the Hugging Face Hub
    path = snapshot_download(repo_id=DATASET_REPO_ID, repo_type="dataset")
    return path


def get_data(rl_env, path) -> pd.DataFrame:
    """
    Get data from rl_env CSV file, format model and user as clickable links, and return as DataFrame
    """
    csv_path = os.path.join(path, rl_env + ".csv")
    data = pd.read_csv(csv_path)

    # Add clickable links for model and user
    for index, row in data.iterrows():
        data.at[index, "User"] = make_clickable_user(row["User"])
        data.at[index, "Model"] = make_clickable_model(row["Model"])

    return data


def get_last_refresh_time(path) -> str:
    """
    Get the last update time from the last_update.txt file in the dataset path.
    """
    # Path to the last_update.txt file
    update_file_path = os.path.join(path, 'last_update.txt')

    # Check if the file exists
    if os.path.exists(update_file_path):
        # Read the content of the file (the timestamp)
        with open(update_file_path, 'r') as f:
            last_refresh_time = f.read().strip()
        return last_refresh_time
    else:
        # Fallback: If the file is missing, return a default message
        return "Last update time not available"


# Function to refresh the dataset periodically
def refresh_dataset():
    global path_
    path_ = download_leaderboard_dataset()  # Redownload the dataset
    global last_refresh_time
    last_refresh_time = get_last_refresh_time(path_)


# Set up a background scheduler to refresh the dataset every hour
scheduler = BackgroundScheduler()
scheduler.add_job(refresh_dataset, 'interval', minutes=15)
scheduler.start()


with block:
    path_ = download_leaderboard_dataset()
    # Get the last refresh time
    last_refresh_time = get_last_refresh_time(path_)

    gr.Markdown(f"""
    # 🏆 Deep Reinforcement Learning Course Leaderboard (Mirror)🏆 

    Presenting the latest leaderboard from the Hugging Face Deep RL Course - refreshed at {last_refresh_time}.
    """)

    for i in range(0, len(rl_envs)):
        rl_env = rl_envs[i]
        with gr.TabItem(rl_env["rl_env_beautiful"]):
            with gr.Row():
                markdown = f"""
                    # {rl_env['rl_env_beautiful']}

                    ### Leaderboard for {rl_env['rl_env_beautiful']}
                    """
                gr.Markdown(markdown)

            with gr.Row():
                # Display the data for this RL environment
                data = get_data(rl_env["rl_env"], path_)
                gr.Dataframe(
                    value=data,
                    headers=["Ranking 🏆", "User 🤗", "Model id 🤖", "Results", "Mean Reward", "Std Reward"],
                    datatype=["number", "markdown", "markdown", "number", "number", "number"],
                    row_count=(100, 'fixed')
                )

block.launch()