app.py

import pathlib
from pathlib import Path
import tempfile
from typing import BinaryIO, Literal
import json
import pandas as pd

import gradio as gr
from datasets import load_dataset
from gradio_leaderboard import ColumnFilter, Leaderboard, SelectColumns
from evaluation import evaluate_problem
from datetime import datetime
import os

from submit import submit_boundary
from about import PROBLEM_TYPES, TOKEN, CACHE_PATH, API, submissions_repo, results_repo
from utils import read_boundary, write_results, get_user
from visualize import make_visual

def evaluate_boundary(filename):
    print(filename)
    local_path = read_boundary(filename)
    with Path(local_path).open("r") as f:
        raw = f.read()
        data_dict = json.loads(raw)
    result = evaluate_problem(data_dict['problem_type'], local_path)
    
    write_results(data_dict, result)
    return

def make_clickable(name):
    link =f'https://huggingface.co/{name}'
    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{name}</a>'

def get_leaderboard():
    ds = load_dataset(results_repo, split='train')
    df = pd.DataFrame(ds)

    df.rename(columns={'submission_time': 'submission time', 'problem_type': 'problem type'}, inplace=True)
    # df['user'] = df['user'].apply(lambda x: make_clickable(x)).astype(str)
    score_field = "score" if "score" in df.columns else "objective"  # fallback

    df = df.sort_values(by=score_field, ascending=True)
    return df

def show_output_box(message):
    return gr.update(value=message, visible=True)

def gradio_interface() -> gr.Blocks:
    with gr.Blocks() as demo:
        gr.Markdown("## Welcome to the ConStellaration Boundary Leaderboard!")
        with gr.Tabs(elem_classes="tab-buttons"):
            with gr.TabItem("🚀 Leaderboard", elem_id="boundary-benchmark-tab-table"):                
                gr.Markdown("# Boundary Design Leaderboard")

                Leaderboard(
                    value=get_leaderboard(),
                    datatype=['str', 'date', 'str', 'str', 'bool', 'markdown', 'number', 'bool', 'number', 'number', 'str'],
                    select_columns=["submission time", "feasibility", "score", "problem type", "user"],
                    search_columns=["submission time", "score", "user"],
                    hide_columns=["result_filename", "submission_filename", "objective", "minimize_objective", "boundary_json", "evaluated"],
                    filter_columns=["problem type"],
                    every=60,
                    render=True
                )

            with gr.TabItem("❔About", elem_id="boundary-benchmark-tab-table"):
                gr.Markdown(
                    """
                    ## About This Challenge

                    **Welcome to the ConStellaration Leaderboard**, a community-driven effort to accelerate fusion energy research using machine learning.

                    In collaboration with [Proxima Fusion](https://www.proximafusion.com/), we're inviting the ML and physics communities to optimize plasma configurations for stellarators—a class of fusion reactors that offer steady-state operation and strong stability advantages over tokamaks.

                    This leaderboard tracks submissions to a series of open benchmark tasks focused on:

                    - **Geometrically optimized stellarators**
                    - **Simple-to-build quasi-isodynamic (QI) stellarators**
                    - **Multi-objective, MHD-stable QI stellarators**

                    Participants are encouraged to build surrogate models, optimize plasma boundaries, and explore differentiable design pipelines that could replace or accelerate slow traditional solvers like VMEC++.

                    ### Why It Matters

                    Fusion promises clean, abundant, zero-carbon energy. But designing stellarators is computationally intense and geometrically complex. With open datasets, reference baselines, and your contributions, we can reimagine this process as fast, iterative, and ML-native.

                    ### How to Participate

                    - Clone the [ConStellaration dataset](https://huggingface.co/datasets/proxima-fusion/constellaration)
                    - Build or train your model on the provided QI equilibria
                    - Submit your predicted boundaries and results here to benchmark against others
                    - Join the discussion and help expand the frontier of fusion optimization

                    Let's bring fusion down to Earth—together.

                """
                )

                # dropdown = gr.Dropdown(choices=filenames, label="Choose a file")
                # plot_output = gr.Plot()

            with gr.TabItem("✉️ Submit", elem_id="boundary-benchmark-tab-table"):
                gr.Markdown(
                    """
                # Plasma Boundary Evaluation Submission
                Upload your plasma boundary JSON and select the problem type to get your score.
                """
                )
                user_state = gr.State(value=None) 
                filename = gr.State(value=None) 
                eval_state = gr.State(value=None) 

                gr.LoginButton()

                demo.load(get_user, inputs=None, outputs=user_state)

                with gr.Row():
                    problem_type = gr.Dropdown(PROBLEM_TYPES, label="Problem Type")
                    boundary_file = gr.File(label="Boundary JSON File (.json)")

                boundary_file
                submit_btn = gr.Button("Evaluate")
                message = gr.Textbox(label="Status", lines=1, visible=False)
                submit_btn.click(
                    submit_boundary,
                    inputs=[problem_type, boundary_file, user_state],
                    outputs=[message, filename],
                ).then(
                    fn=show_output_box,
                    inputs=[message],
                    outputs=[message],
                ).then(
                    fn=evaluate_boundary,
                    inputs=[filename],
                    outputs=[eval_state]
                )
 
    return demo


if __name__ == "__main__":
    gradio_interface().launch()