eval_check.py

Go to the documentation of this file.

# Copyright (c) 2025-present Aquin Labs Private Limited. All Rights Reserved.
"""`aquin eval custom --check` — save JSON + diagram to the current working directory."""
 
from __future__ import annotations
 
import json
from datetime import datetime, timezone
from pathlib import Path
from typing import Any
 
_JSON_NAME = "eval-check.json"
_PNG_NAME = "eval-check.png"
 
 
def _normalize_eval_result(result: dict[str, Any]) -> dict[str, Any]:
    if not isinstance(result, dict):
        return {"error": "Invalid eval result"}
 
    if result.get("error"):
        return dict(result)
 
    data = result
    if isinstance(result.get("content"), dict):
        data = result["content"]
    card = result.get("card")
    if isinstance(card, dict) and isinstance(card.get("data"), dict):
        data = card["data"]
 
    return dict(data)
 
 
def has_eval_payload(result: dict[str, Any]) -> bool:
    data = _normalize_eval_result(result)
    if data.get("error"):
        return False
    return bool(data.get("prompts")) or data.get("mean_score") is not None
 
 
def write_eval_check(
    result: dict[str, Any],
    *,
    tool_name: str | None,
    cwd: str | Path,
) -> tuple[Path, Path]:
    cwd = Path(cwd)
    json_path = cwd / _JSON_NAME
    png_path = cwd / _PNG_NAME
 
    flat = _normalize_eval_result(result)
    payload = {
        "saved_at": datetime.now(timezone.utc).isoformat(),
        "tool": tool_name,
        **flat,
    }
    json_path.write_text(json.dumps(payload, indent=2, default=str), encoding="utf-8")
    _plot_eval_check(flat, png_path)
    return json_path, png_path
 
 
def _plot_eval_check(result: dict[str, Any], png_path: Path) -> None:
    import matplotlib
 
    matplotlib.use("Agg")
    import matplotlib.pyplot as plt
 
    if result.get("error"):
        fig, ax = plt.subplots(figsize=(6, 2), facecolor="#0f1117")
        ax.axis("off")
        ax.text(0.5, 0.5, f"Error: {result['error']}", ha="center", va="center", wrap=True, color="#e5e7eb")
        fig.savefig(png_path, dpi=140, bbox_inches="tight", facecolor="#0f1117")
        plt.close(fig)
        return
 
    name = str(result.get("name") or "eval")
    model = str(result.get("model_id") or "")
    threshold = float(result.get("threshold") or 0.5)
    mean_score = float(result.get("mean_score") or 0)
    pass_rate = float(result.get("pass_rate") or 0)
    rows = result.get("prompts") or []
 
    title = f"Custom eval — {name}"
    if model:
        title += f"  ·  {model}"
    subtitle = f"mean {mean_score:.0%}  ·  pass rate {pass_rate:.0%}  ·  threshold {threshold:.0%}"
 
    fig, axes = plt.subplots(1, 2, figsize=(11, max(4.5, min(12, len(rows) * 0.38 + 2))), facecolor="#0f1117")
    fig.suptitle(title, color="#e5e7eb", fontsize=11, y=1.02)
    fig.text(0.5, 0.96, subtitle, ha="center", fontsize=8, color="#9ca3af")
 
    _plot_prompt_scores(axes[0], rows, threshold)
    _plot_summary_panel(axes[1], mean_score, pass_rate, threshold, rows)
 
    for ax in axes:
        ax.set_facecolor("#0f1117")
        for spine in ax.spines.values():
            spine.set_color("#374151")
 
    fig.tight_layout(rect=(0, 0, 1, 0.93))
    fig.savefig(png_path, dpi=140, bbox_inches="tight", facecolor="#0f1117")
    plt.close(fig)
 
 
def _plot_prompt_scores(ax: Any, rows: list[dict[str, Any]], threshold: float) -> None:
    ax.set_title("Per-prompt score", color="#e5e7eb", fontsize=10)
 
    if not rows:
        ax.axis("off")
        ax.text(0.5, 0.5, "No prompts", ha="center", va="center", color="#9ca3af")
        return
 
    labels = [_trunc(str(r.get("prompt") or ""), 28) for r in rows]
    vals = [float(r.get("score") or 0) for r in rows]
    colors = ["#34d399" if r.get("passed") else "#f87171" for r in rows]
 
    y_pos = list(range(len(rows)))
    ax.barh(y_pos, vals, color=colors, height=0.65, alpha=0.9)
    ax.axvline(threshold, color="#fbbf24", linestyle="--", linewidth=0.8, alpha=0.75)
    ax.set_yticks(y_pos)
    ax.set_yticklabels(labels, fontsize=7, color="#d1d5db")
    ax.invert_yaxis()
    ax.set_xlim(0, 1.05)
    ax.set_xlabel("Score (0–1)", color="#9ca3af", fontsize=8)
    ax.tick_params(axis="x", colors="#6b7280", labelsize=8)
    ax.grid(axis="x", alpha=0.2, color="#4b5563")
    for i, v in enumerate(vals):
        ax.text(min(v + 0.02, 0.98), i, f"{v:.2f}", va="center", fontsize=7, color="#9ca3af")
 
 
def _plot_summary_panel(
    ax: Any,
    mean_score: float,
    pass_rate: float,
    threshold: float,
    rows: list[dict[str, Any]],
) -> None:
    ax.set_title("Summary", color="#e5e7eb", fontsize=10)
 
    metrics = [("mean score", mean_score), ("pass rate", pass_rate), ("threshold", threshold)]
    labels = [m[0] for m in metrics]
    vals = [m[1] for m in metrics]
    colors = [_score_color(mean_score), _score_color(pass_rate), "#78716c"]
 
    y_pos = list(range(len(metrics)))
    ax.barh(y_pos, vals, color=colors, height=0.55, alpha=0.9)
    ax.set_yticks(y_pos)
    ax.set_yticklabels(labels, fontsize=9, color="#d1d5db")
    ax.invert_yaxis()
    ax.set_xlim(0, 1.05)
    ax.set_xlabel("Value (0–1)", color="#9ca3af", fontsize=8)
    ax.tick_params(axis="x", colors="#6b7280", labelsize=8)
    ax.grid(axis="x", alpha=0.2, color="#4b5563")
    for i, v in enumerate(vals):
        ax.text(min(v + 0.02, 0.98), i, f"{v:.0%}", va="center", fontsize=8, color="#9ca3af")
 
    n_pass = sum(1 for r in rows if r.get("passed"))
    ax.text(
        0.02, 0.02,
        f"{n_pass}/{len(rows)} passed",
        transform=ax.transAxes,
        fontsize=8,
        color="#9ca3af",
    )
 
 
def _score_color(val: float) -> str:
    if val >= 0.75:
        return "#34d399"
    if val >= 0.5:
        return "#fbbf24"
    return "#f87171"
 
 
def _trunc(text: str, limit: int) -> str:
    t = text.strip()
    if len(t) <= limit:
        return t or "·"
    return t[: limit - 1] + "…"