src/populate.py

import json
import os

import pandas as pd

from src.display.formatting import has_no_nan_values, make_clickable_model
from src.display.utils import AutoEvalColumn, EvalQueueColumn
from src.leaderboard.read_evals import get_raw_eval_results


def get_leaderboard_df(results_path):
    df = pd.read_csv(results_path)
    # numeric formatting
    df["ha_rag_rate"]  = df["ha_rag_rate"].round(2)
    df["ha_non_rag_rate"] = df["ha_non_rag_rate"].round(2)

    # --- map to pretty headers just before returning ---
    pretty = {
        "Models":                 "Models",
        "ha_rag_rate":     "RAG Hallucination Rate (%)",
        "ha_non_rag_rate": "Non-RAG Hallucination Rate (%)",
    }
    df = df.rename(columns=pretty) # this is what the UI will use
    # ----------- Average column & ranking ---------------------------------------------
    df["Average Hallucination Rate (%)"] = df[
        ["RAG Hallucination Rate (%)", "Non-RAG Hallucination Rate (%)"]
    ].mean(axis=1).round(2)

    # sort so *lower* average = better (true leaderboard style)
    df = df.sort_values("Average Hallucination Rate (%)", ascending=True).reset_index(drop=True)

    # # Rank & medal
    medal_map  = {1: "🥇", 2: "🥈", 3: "🥉"} 

    def medal_html(rank):
        """Return an HTML span with the medal icon for the top 3 ranks.

        The numeric rank is stored in the data-order attribute equal to the numerical rank so that
        DataTables (used under-the-hood by the gradio_leaderboard component)
        can sort the column by this hidden numeric value while still
        displaying the pretty medal icon. For ranks > 3 we just return the
        integer so the column remains fully numeric.
        """
        medal = medal_map.get(rank)
        if medal:
            # Prepend a hidden numeric span so string sorting still works numerically.
            return (
                f'<span style="display:none">{rank:04}</span>'  # zero-padded for stable string sort
                f'<span style="font-size:2.0rem;">{medal}</span>'
            )
        # For other ranks, also zero-pad to keep width and ensure proper string sort
        return f'<span style="display:none">{rank:04}</span>{rank}'
    
    df["Rank"] = df.index + 1
    df["Rank"] = df["Rank"].apply(medal_html)


    # ----------- column ordering ------------------------------------------------------
    df = df[[
        "Rank",                 # pretty column user sees
        "Models",
        "Average Hallucination Rate (%)",
        "RAG Hallucination Rate (%)",
        "Non-RAG Hallucination Rate (%)",
    ]]

    return df  




def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:
    """Creates the different dataframes for the evaluation queues requestes"""
    entries = [entry for entry in os.listdir(save_path) if not entry.startswith(".")]
    all_evals = []

    for entry in entries:
        if ".json" in entry:
            file_path = os.path.join(save_path, entry)
            with open(file_path) as fp:
                data = json.load(fp)

            data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
            data[EvalQueueColumn.revision.name] = data.get("revision", "main")

            all_evals.append(data)
        elif ".md" not in entry:
            # this is a folder
            sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if os.path.isfile(e) and not e.startswith(".")]
            for sub_entry in sub_entries:
                file_path = os.path.join(save_path, entry, sub_entry)
                with open(file_path) as fp:
                    data = json.load(fp)

                data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
                data[EvalQueueColumn.revision.name] = data.get("revision", "main")
                all_evals.append(data)

    pending_list = [e for e in all_evals if e["status"] in ["PENDING", "RERUN"]]
    running_list = [e for e in all_evals if e["status"] == "RUNNING"]
    finished_list = [e for e in all_evals if e["status"].startswith("FINISHED") or e["status"] == "PENDING_NEW_EVAL"]
    df_pending = pd.DataFrame.from_records(pending_list, columns=cols)
    df_running = pd.DataFrame.from_records(running_list, columns=cols)
    df_finished = pd.DataFrame.from_records(finished_list, columns=cols)
    return df_finished[cols], df_running[cols], df_pending[cols]