core/comparison_table.py

"""
🇬🇧 Module: comparison_table.py
Purpose: Generates comparative DataFrame for two portfolios and an LLM commentary.

🇷🇺 Модуль: comparison_table.py
Назначение: создаёт сравнительную таблицу метрик двух портфелей и комментарий LLM.
"""

import pandas as pd
import asyncio
from services.output_api import fetch_metrics_async, extract_portfolio_id
from services.llm_client import llm_service
from prompts.system_prompts import COMPARISON_SYSTEM_PROMPT


def show_comparison_table(portfolio_a: str, portfolio_b: str):
    """Public Gradio entry: returns both a DataFrame and LLM commentary."""
    pid_a = extract_portfolio_id(portfolio_a)
    pid_b = extract_portfolio_id(portfolio_b)
    if not pid_a or not pid_b:
        return "❌ Invalid portfolio IDs.", "No commentary available."

    try:
        df, commentary = asyncio.run(_build_comparison_with_comment(pid_a, pid_b))
        return df, commentary
    except Exception as e:
        return f"❌ Error building comparison table: {e}", "❌ LLM analysis failed."


async def _build_comparison_with_comment(p1: str, p2: str):
    """Async helper: builds table and gets commentary."""
    m1 = await fetch_metrics_async(p1)
    m2 = await fetch_metrics_async(p2)
    if not m1 or not m2:
        raise ValueError("Metrics unavailable for one or both portfolios.")

    all_keys = sorted(set(m1.keys()) | set(m2.keys()))
    rows = []
    for k in all_keys:
        v1 = m1.get(k, 0)
        v2 = m2.get(k, 0)
        diff = v1 - v2
        symbol = "▲" if diff > 0 else "▼" if diff < 0 else "—"
        rows.append({
            "Metric": k,
            "Portfolio A": round(v1, 3),
            "Portfolio B": round(v2, 3),
            "Δ Difference": f"{symbol} {diff:+.3f}"
        })
    df = pd.DataFrame(rows, columns=["Metric", "Portfolio A", "Portfolio B", "Δ Difference"])

    # Generate LLM commentary
    summary = "\n".join(f"{r['Metric']}: {r['Δ Difference']}" for r in rows)
    prompt = (
        f"{COMPARISON_SYSTEM_PROMPT}\n"
        f"Compare and explain the differences between Portfolio A and B:\n{summary}\n"
        f"Write your insights as a concise professional commentary."
    )

    commentary = ""
    for delta in llm_service.stream_chat(
        messages=[
            {"role": "system", "content": "You are an investment portfolio analyst."},
            {"role": "user", "content": prompt},
        ],
        model="meta-llama/Meta-Llama-3.1-8B-Instruct",
    ):
        commentary += delta

    return df, commentary