suppression_eval_check.py

Go to the documentation of this file.

# Copyright (c) 2025-present Aquin Labs Private Limited. All Rights Reserved.
"""`aquin suppression-eval --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 = "suppression-eval-check.json"
_PNG_NAME = "suppression-eval-check.png"
 
_STATUS_COLORS = {
    "suppressed": "#f87171",
    "softened": "#fbbf24",
    "unfiltered": "#34d399",
}
 
 
def _normalize_suppression_eval_result(result: dict[str, Any]) -> dict[str, Any]:
    if not isinstance(result, dict):
        return {"error": "Invalid suppression-eval result"}
 
    if result.get("error"):
        return dict(result)
 
    data = result
    if isinstance(result.get("content"), dict):
        data = result["content"]
    if isinstance(data.get("suppression"), dict):
        data = data["suppression"]
    card = result.get("card")
    if isinstance(card, dict) and isinstance(card.get("data"), dict):
        nested = card["data"].get("suppression")
        if isinstance(nested, dict):
            data = nested
 
    return dict(data)
 
 
def has_suppression_eval_payload(result: dict[str, Any]) -> bool:
    data = _normalize_suppression_eval_result(result)
    if data.get("error"):
        return False
    return bool(data.get("topics")) or bool(data.get("baseline"))
 
 
def write_suppression_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_suppression_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_suppression_eval_check(flat, png_path)
    return json_path, png_path
 
 
def _plot_suppression_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
 
    baseline = result.get("baseline") or {}
    topics = result.get("topics") or []
    n_topics = len(topics)
    suppressed = sum(1 for t in topics if t.get("status") == "suppressed")
    softened = sum(1 for t in topics if t.get("status") == "softened")
 
    title = f"Suppression eval — {n_topics} topics"
    subtitle = f"baseline len {baseline.get('mean_length', '—')}  ·  hedge {baseline.get('mean_hedge', '—')}  ·  {suppressed} suppressed  ·  {softened} softened"
 
    fig, axes = plt.subplots(1, 2, figsize=(11, max(4.5, min(10, n_topics * 0.35 + 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_suppression_scores(axes[0], topics)
    _plot_ratio_panel(axes[1], topics)
 
    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_suppression_scores(ax: Any, topics: list[dict[str, Any]]) -> None:
    ax.set_title("Suppression score by topic", color="#e5e7eb", fontsize=10)
 
    if not topics:
        ax.axis("off")
        ax.text(0.5, 0.5, "No topics", ha="center", va="center", color="#9ca3af")
        return
 
    rows = sorted(topics, key=lambda t: float(t.get("suppression_score") or 0), reverse=True)
    labels = [str(t.get("topic") or "?") for t in rows]
    vals = [float(t.get("suppression_score") or 0) for t in rows]
    colors = [_STATUS_COLORS.get(str(t.get("status") or ""), "#78716c") for t in rows]
 
    y_pos = list(range(len(rows)))
    ax.barh(y_pos, vals, color=colors, height=0.65, alpha=0.9)
    ax.set_yticks(y_pos)
    ax.set_yticklabels(labels, fontsize=8, color="#d1d5db")
    ax.invert_yaxis()
    ax.set_xlim(0, 1.05)
    ax.set_xlabel("Suppression score", 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_ratio_panel(ax: Any, topics: list[dict[str, Any]]) -> None:
    ax.set_title("Length × hedge vs baseline", color="#e5e7eb", fontsize=10)
 
    if not topics:
        ax.axis("off")
        ax.text(0.5, 0.5, "No topics", ha="center", va="center", color="#9ca3af")
        return
 
    labels = [str(t.get("topic") or "?")[:12] for t in topics]
    x = list(range(len(topics)))
    width = 0.38
    len_ratios = [float(t.get("length_ratio") or 0) for t in topics]
    hedge_ratios = [float(t.get("hedge_ratio") or 0) for t in topics]
 
    ax.bar([i - width / 2 for i in x], len_ratios, width=width, color="#6366f1", alpha=0.9, label="length×")
    ax.bar([i + width / 2 for i in x], hedge_ratios, width=width, color="#fbbf24", alpha=0.9, label="hedge×")
    ax.axhline(1.0, color="#9ca3af", linestyle="--", linewidth=0.8, alpha=0.5)
    ax.set_xticks(x)
    ax.set_xticklabels(labels, fontsize=7, color="#d1d5db", rotation=30, ha="right")
    ax.set_ylabel("Ratio vs baseline", color="#9ca3af", fontsize=8)
    ax.legend(fontsize=7, facecolor="#0f1117", edgecolor="#374151", labelcolor="#d1d5db")
    ax.tick_params(colors="#6b7280", labelsize=7)
    ax.grid(axis="y", alpha=0.2, color="#4b5563")