app.py

from pathlib import Path
import json
import pandas as pd

import gradio as gr
from gradio_leaderboard import Leaderboard

from utils import read_submission_from_hub, write_results
from about import ASSAY_LIST, ASSAY_RENAME, ASSAY_EMOJIS, submissions_repo, API, results_repo
from typing import BinaryIO, Literal
from datetime import datetime
import tempfile
from datasets import load_dataset
import io

def make_submission(
    submitted_file: BinaryIO,
    user_state):

    if user_state is None:
        raise gr.Error("You must submit your username to submit a file.")
    
    file_path = submitted_file.name

    if not file_path:
        raise gr.Error("Uploaded file object does not have a valid file path.")
    
    path_obj = Path(file_path)
    timestamp = datetime.utcnow().isoformat()

    with (path_obj.open("rb") as f_in):
        file_content = f_in.read().decode("utf-8")

        # write to dataset
        filename = f"{user_state}/{timestamp.replace(':', '-')}_{user_state}.json"
        record = {
            "submission_filename": filename,
            "submission_time": timestamp,
            "csv_content": file_content,
            "evaluated": False,
            "user": user_state,
        }
        with tempfile.NamedTemporaryFile(mode="w", suffix=".json", delete=False) as tmp:
            json.dump(record, tmp, indent=2)
            tmp.flush()
            tmp_name = tmp.name

        API.upload_file(
            path_or_fileobj=tmp_name, 
            path_in_repo=filename,
            repo_id=submissions_repo,
            repo_type="dataset",
            commit_message=f"Add submission for {user_state} at {timestamp}"
        )
        Path(tmp_name).unlink()

    return "✅ Your submission has been received! Sit tight and your scores will appear on the leaderboard shortly."

    
def get_leaderboard_table(df_results: pd.DataFrame, assay: str | None = None):
    # ds = load_dataset(results_repo, split='train', download_mode="force_redownload")
    # full_df = pd.DataFrame(ds)
    # full_df['full results'] = full_df['result_filename'].apply(lambda x: make_boundary_clickable(x)).astype(str)

    # full_df.rename(columns={'submission_time': 'submission time', 'problem_type': 'problem type'}, inplace=True)
    # to_show = full_df.copy(deep=True)
    # to_show = to_show[to_show['user'] != 'test']
    # to_show = to_show[['submission time', 'problem type', 'user', 'score', 'full results']]
    # to_show['user'] = to_show['user'].apply(lambda x: make_user_clickable(x)).astype(str)
    
    # Previously hosted on HF hub, local for now (Can also pull directly from github backend)
    column_order = ["model", "property", "spearman", "spearman_abs"]
    df = df_results.query("assay.isin(@ASSAY_RENAME.keys())").copy()
    if assay is not None:
        df = df[df['assay'] == assay]
    df = df[column_order]
    return df.sort_values(by="spearman_abs", ascending=False)

def get_leaderboard_object(df_results: pd.DataFrame, assay: str | None = None):
    df = get_leaderboard_table(df_results=df_results, assay=assay)
    filter_columns = ["model"]
    if assay is None:
        filter_columns.append("property")
    # TODO how to sort filter columns alphabetically?
    Leaderboard(
        value=df,
        datatype=["str", "str", "str", "number"],
        select_columns=["model", "property", "spearman"],
        search_columns=["model"],
        filter_columns=filter_columns,
        every=60,
        render=True
    )

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

def fetch_hf_results():
    ds = load_dataset(results_repo, split='no_low_spearman', download_mode="force_redownload")
    df = pd.DataFrame(ds).drop_duplicates(subset=["model", "assay"])
    df["property"] = df["assay"].map(ASSAY_RENAME)
    return df

with gr.Blocks() as demo:
    gr.Markdown("""
        ## Welcome to the Ginkgo Antibody Developability Benchmark Leaderboard!
        
        Participants can submit their model to the leaderboard by 
        """)
    df = fetch_hf_results()
    with gr.Tabs(elem_classes="tab-buttons"):
        # Procedurally make these 5 tabs
        for assay in ASSAY_LIST:
            with gr.TabItem(f"{ASSAY_EMOJIS[assay]} {ASSAY_RENAME[assay]}", elem_id=f"abdev-benchmark-tab-table"):
                gr.Markdown(f"# {ASSAY_RENAME[assay]} (measured by {assay})")
                get_leaderboard_object(df_results=df, assay=assay)
        
        with gr.TabItem("🚀 Overall", elem_id="abdev-benchmark-tab-table"):                
            gr.Markdown("# Antibody Developability Benchmark Leaderboard over all properties")

            get_leaderboard_object(df_results=df)
            # TODO: this is not going to update well, need to fix
        
        with gr.TabItem("❔About", elem_id="abdev-benchmark-tab-table"):
            gr.Markdown(
                """
                ## About this challenge
                
                We're inviting the ML/bio community to predict developability properties for 244 antibodies from the [GDPa1 dataset](https://huggingface.co/datasets/ginkgo-datapoints/GDPa1).
                
                **What is antibody developability?**
                
                Antibodies have to be manufacturable, stable in high concentrations, and have low off-target effects. 
                Properties such as these can often hinder the progression of an antibody to the clinic, and are collectively referred to as 'developability'.
                Here we show 5 of these properties and invite the community to submit and develop better predictors, which will be tested out on a heldout private set to assess model generalization.
                
                **How to submit?**
                
                TODO
                
                **How to evaluate?**
                
                TODO
            """
            )

        with gr.TabItem("✉️ Submit", elem_id="boundary-benchmark-tab-table"):
                gr.Markdown(
                    """
                # Antibody Developability Submission
                Upload a CSV to get a score!
                """
                )
                filename = gr.State(value=None) 
                eval_state = gr.State(value=None) 
                user_state = gr.State(value=None)

                # gr.LoginButton()

                with gr.Row():
                    with gr.Column():
                        username_input = gr.Textbox(
                            label="Username", 
                            placeholder="Enter your Hugging Face username",
                            info="This will be displayed on the leaderboard."
                        )
                    with gr.Column():
                        boundary_file = gr.File(label="Submission CSV")

                username_input.change(
                    fn=lambda x: x if x.strip() else None,
                    inputs=username_input,
                    outputs=user_state
                )                       

                submit_btn = gr.Button("Evaluate")
                message = gr.Textbox(label="Status", lines=1, visible=False)
                # help message
                gr.Markdown("If you have issues with submission or using the leaderboard, please start a discussion in the Community tab of this Space.")
                
                submit_btn.click(
                    make_submission,
                    inputs=[boundary_file, user_state],
                    outputs=[message],
                ).then(
                    fn=show_output_box,
                    inputs=[message],
                    outputs=[message],
                )


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