Agent/examples_how/overall_derivation/tools/find_tree_node.py

"""
查找树节点 Tool - 人设树节点查询

功能：
1. 获取人设树的常量节点（全局常量、局部常量）
2. 获取符合条件概率阈值的节点（按条件概率排序返回 topN）
"""

import json
import sys
from pathlib import Path
from typing import Any, Optional

# 保证直接运行或作为包加载时都能解析 utils / tools（IDE 可跳转）
_root = Path(__file__).resolve().parent.parent
if str(_root) not in sys.path:
    sys.path.insert(0, str(_root))
from utils.conditional_ratio_calc import calc_node_conditional_ratio  # noqa: E402
from tools.point_match import match_derivation_to_post_points  # noqa: E402

try:
    from agent.tools import tool, ToolResult, ToolContext
except ImportError:
    def tool(*args, **kwargs):
        return lambda f: f
    ToolResult = None  # 仅用 main() 测核心逻辑时可无 agent
    ToolContext = None

# 相对本文件：tools -> overall_derivation，input / output 在 overall_derivation 下
_BASE_INPUT = Path(__file__).resolve().parent.parent / "input"
_BASE_OUTPUT = Path(__file__).resolve().parent.parent / "output"


def _dimension_analysis_log_dir(account_name: str, post_id: str, log_id: str) -> Path:
    """推导日志目录：output/{account_name}/推导日志/{post_id}/{log_id}/"""
    return _BASE_OUTPUT / account_name / "推导日志" / post_id / log_id


def _load_derived_dim_tree_node_names(
    account_name: str, post_id: str, log_id: str, round: int
) -> list[str]:
    """
    读取当前轮次对应的维度分析 JSON（优先 {round}_维度分析.json，不存在则 {round-1}_维度分析.json），
    返回 derived_dims 中每项的 tree_node_name（已推导出的维度节点，人设树中层次较高）。
    无可用文件时返回空列表。
    """
    if not log_id or not str(log_id).strip():
        return []
    log_dir = _dimension_analysis_log_dir(account_name, post_id, str(log_id).strip())
    for r in (round, round - 1):
        if r < 1:
            continue
        path = log_dir / f"{r}_维度分析.json"
        if not path.is_file():
            continue
        try:
            with open(path, "r", encoding="utf-8") as f:
                data = json.load(f)
        except Exception:
            continue
        dims = data.get("derived_dims") or []
        names: list[str] = []
        for d in dims:
            if isinstance(d, dict):
                tn = d.get("tree_node_name")
                if tn is not None and str(tn).strip():
                    names.append(str(tn).strip())
        return names
    return []


def _descendant_names_under_tree_nodes(
    account_name: str, anchor_node_names: list[str]
) -> tuple[set[str], dict[str, str]]:
    """
    在每个人设维度树根上 DFS，收集所有锚点（derived_dims.tree_node_name）之下的**全部后代**（不含锚点自身）。

    同时记录「所属维度」：对路径上每个后代节点，取从维度根到该节点路径上**最深的**那个锚点
    （与原先沿父链向上找最近 derived_dim 一致；多个锚点呈祖孙时取更深者）。

    Returns:
        (allowed 节点名集合, 节点名 -> 所属已推导维度树节点名)
    """
    if not anchor_node_names:
        return set(), {}
    S = set(anchor_node_names)
    allowed: set[str] = set()
    dim_map: dict[str, str] = {}

    for dim_root_name, root in _load_trees(account_name):

        def dfs(node_name: str, node_dict: dict, parent_deepest_s: Optional[str]) -> None:
            d_self = node_name if node_name in S else parent_deepest_s
            for cname, cnode in (node_dict.get("children") or {}).items():
                if not isinstance(cnode, dict):
                    continue
                if cname not in S and d_self is not None:
                    allowed.add(cname)
                    dim_map[cname] = d_self
                dfs(cname, cnode, d_self)

        dfs(dim_root_name, root, None)

    return allowed, dim_map


def _tree_dir(account_name: str) -> Path:
    """人设树目录：../input/{account_name}/原始数据/tree/"""
    return _BASE_INPUT / account_name / "原始数据" / "tree"


def _load_trees(account_name: str) -> list[tuple[str, dict]]:
    """加载该账号下所有维度的人设树。返回 [(维度名, 根节点 dict), ...]。"""
    td = _tree_dir(account_name)
    if not td.is_dir():
        return []
    result = []
    for p in td.glob("*.json"):
        try:
            with open(p, "r", encoding="utf-8") as f:
                data = json.load(f)
            for dim_name, root in data.items():
                if isinstance(root, dict):
                    result.append((dim_name, root))
                break
        except Exception:
            continue
    return result


def _iter_all_nodes(account_name: str):
    """遍历该账号下所有人设树节点，产出 (节点名称, 父节点名称, 节点 dict)。"""
    for dim_name, root in _load_trees(account_name):
        def walk(parent_name: str, node_dict: dict):
            for name, child in (node_dict.get("children") or {}).items():
                if not isinstance(child, dict):
                    continue
                yield (name, parent_name, child)
                yield from walk(name, child)

        yield from walk(dim_name, root)


# ---------------------------------------------------------------------------
# 1. 获取人设树常量节点
# ---------------------------------------------------------------------------

def get_constant_nodes(account_name: str) -> list[dict[str, Any]]:
    """
    获取人设树的常量节点。
    - 全局常量：_is_constant=True
    - 局部常量：_is_local_constant=True 且 _is_constant=False
    返回列表项：节点名称、概率(_ratio)、常量类型。
    """
    result = []
    for node_name, _parent, node in _iter_all_nodes(account_name):
        is_const = node.get("_is_constant") is True
        is_local = node.get("_is_local_constant") is True
        if is_const:
            const_type = "全局常量"
        elif is_local and not is_const:
            const_type = "局部常量"
        else:
            continue
        ratio = node.get("_ratio")
        result.append({
            "节点名称": node_name,
            "概率": ratio,
            "常量类型": const_type,
        })
    result.sort(key=lambda x: (x["概率"] is None, -(x["概率"] or 0)))
    return result


# ---------------------------------------------------------------------------
# 2. 获取符合条件概率阈值的节点
# ---------------------------------------------------------------------------

def get_nodes_by_conditional_ratio(
    account_name: str,
    derived_list: list[tuple[str, str]],
    threshold: float,
    top_n: int,
    allowed_node_names: Optional[set[str]] = None,
    node_belonging_dim: Optional[dict[str, str]] = None,
) -> list[dict[str, Any]]:
    """
    获取人设树中条件概率 >= threshold 的节点，按条件概率降序，返回前 top_n 个。
    derived_list: 已推导列表，每项 (已推导的选题点, 推导来源人设树节点)；为空时使用节点自身的 _ratio 作为条件概率。
    allowed_node_names: 若给定，仅保留节点名称在该集合内的结果。
    node_belonging_dim: 与 allowed 同步生成（见 _descendant_names_under_tree_nodes），节点名 -> 所属已推导维度；不传则所属维度均为「—」。
    返回列表项：节点名称、条件概率、父节点名称、所属维度。
    """
    base_dir = _BASE_INPUT
    node_to_parent: dict[str, str] = {}
    if derived_list:
        for n, p, _ in _iter_all_nodes(account_name):
            node_to_parent[n] = p

    def dim_for(node_name: str) -> str:
        if not node_belonging_dim:
            return "—"
        return node_belonging_dim.get(node_name) or "—"

    scored: list[tuple[str, float, str, str]] = []

    if not derived_list:
        for node_name, parent_name, node in _iter_all_nodes(account_name):
            if allowed_node_names is not None and node_name not in allowed_node_names:
                continue
            ratio = node.get("_ratio")
            if ratio is None:
                ratio = 0.0
            else:
                ratio = float(ratio)
            if ratio >= threshold:
                scored.append((node_name, ratio, parent_name, dim_for(node_name)))
    else:
        for node_name, parent_name in node_to_parent.items():
            if allowed_node_names is not None and node_name not in allowed_node_names:
                continue
            ratio = calc_node_conditional_ratio(
                account_name, derived_list, node_name, base_dir=base_dir
            )
            if ratio >= threshold:
                scored.append((node_name, ratio, parent_name, dim_for(node_name)))

    scored.sort(key=lambda x: x[1], reverse=True)
    top = scored[:top_n]
    return [
        {
            "节点名称": name,
            "条件概率": ratio,
            "父节点名称": parent,
            "所属维度": dim,
        }
        for name, ratio, parent, dim in top
    ]


def _parse_derived_list(derived_items: list[dict[str, str]]) -> list[tuple[str, str]]:
    """将 agent 传入的 [{"topic": "x", "source_node": "y"}, ...] 转为 DerivedItem 列表。"""
    out = []
    for item in derived_items:
        if isinstance(item, dict):
            topic = item.get("topic") or item.get("已推导的选题点")
            source = item.get("source_node") or item.get("推导来源人设树节点")
            if topic is not None and source is not None:
                out.append((str(topic).strip(), str(source).strip()))
        elif isinstance(item, (list, tuple)) and len(item) >= 2:
            out.append((str(item[0]).strip(), str(item[1]).strip()))
    return out


# ---------------------------------------------------------------------------
# Agent Tools（参考 glob_tool 封装）
# ---------------------------------------------------------------------------

@tool()
async def find_tree_constant_nodes(
    account_name: str,
    post_id: str,
) -> ToolResult:
    """
    获取人设树中的常量节点列表（全局常量与局部常量），并检查每个节点与帖子选题点的匹配情况。

    Args:
    account_name : 账号名，用于定位该账号的人设树数据。
    post_id : 帖子ID，用于加载帖子选题点并与各常量节点做匹配判断。

    Returns:
    ToolResult：
        - title: 结果标题。
        - output: 可读的节点列表文本（每行：节点名称、概率、常量类型、帖子匹配情况）。
        - 出错时 error 为错误信息。
    """
    tree_dir = _tree_dir(account_name)
    if not tree_dir.is_dir():
        return ToolResult(
            title="人设树目录不存在",
            output=f"目录不存在: {tree_dir}",
            error="Directory not found",
        )
    try:
        items = get_constant_nodes(account_name)
        # 批量匹配所有节点与帖子选题点
        if items and post_id:
            node_names = [x["节点名称"] for x in items]
            matched_results = await match_derivation_to_post_points(node_names, account_name, post_id)
            node_match_map: dict[str, list] = {}
            for m in matched_results:
                node_match_map.setdefault(m["推导选题点"], []).append({
                    "帖子选题点": m["帖子选题点"],
                    "匹配分数": m["匹配分数"],
                })
            for item in items:
                matches = node_match_map.get(item["节点名称"], [])
                item["帖子选题点匹配"] = matches if matches else "无"
        if not items:
            output = "未找到常量节点"
        else:
            lines = []
            for x in items:
                match_info = x.get("帖子选题点匹配", "无")
                if isinstance(match_info, list):
                    match_str = "、".join(f"{m['帖子选题点']}({m['匹配分数']})" for m in match_info)
                else:
                    match_str = str(match_info)
                lines.append(f"- {x['节点名称']}\t概率={x['概率']}\t{x['常量类型']}\t帖子选题点匹配={match_str}")
            output = "\n".join(lines)
        return ToolResult(
            title=f"常量节点 ({account_name})",
            output=output,
            metadata={"account_name": account_name, "count": len(items)},
        )
    except Exception as e:
        return ToolResult(
            title="获取常量节点失败",
            output=str(e),
            error=str(e),
        )


@tool()
async def find_tree_nodes_by_conditional_ratio(
    account_name: str,
    post_id: str,
    derived_items: list[dict[str, str]],
    conditional_ratio_threshold: float,
    top_n: int = 100,
    round: int = 1,
    log_id: str = "",
) -> ToolResult:
    """
    按条件概率阈值从人设树筛选节点，返回最多 top_n 条（按条件概率降序），并检查每个节点与帖子选题点的匹配情况。

    Args:
    account_name : 账号名，用于定位该账号的人设树数据。
    post_id : 帖子ID，用于加载帖子选题点并与各节点做匹配判断。
    derived_items : 已推导选题点列表，可为空。非空时每项为字典，需含 topic（或「已推导的选题点」）与 source_node（或「推导来源人设树节点」）
    conditional_ratio_threshold : 条件概率阈值，仅返回条件概率 >= 该值的节点。
    top_n : 返回条数上限。
    round : 推导轮次。
    log_id : 推导日志ID

    Returns:
    ToolResult：
        - title: 结果标题。
        - output: 可读的节点列表文本（每行：节点名称、条件概率、父节点、所属维度、帖子匹配情况）。
        - 出错时 error 为错误信息。
    """
    tree_dir = _tree_dir(account_name)
    if not tree_dir.is_dir():
        return ToolResult(
            title="人设树目录不存在",
            output=f"目录不存在: {tree_dir}",
            error="Directory not found",
        )
    try:
        derived_list = _parse_derived_list(derived_items or [])

        allowed: Optional[set[str]] = None
        node_belonging_dim: dict[str, str] = {}
        dim_source = ""
        derived_dim_names: list[str] = []
        if log_id and str(log_id).strip():
            derived_dim_names = _load_derived_dim_tree_node_names(
                account_name, post_id, str(log_id).strip(), int(round)
            )
            if derived_dim_names:
                allowed, node_belonging_dim = _descendant_names_under_tree_nodes(
                    account_name, derived_dim_names
                )
                # 记录实际用到的维度分析文件（与读取逻辑一致）
                log_dir = _dimension_analysis_log_dir(account_name, post_id, str(log_id).strip())
                for r in (int(round), int(round) - 1):
                    if r >= 1 and (log_dir / f"{r}_维度分析.json").is_file():
                        dim_source = f"{r}_维度分析.json (derived_dims -> 全部后代)"
                        break
            else:
                dim_source = "未读到 derived_dims（无对应维度分析文件或为空），未收窄"

        items = get_nodes_by_conditional_ratio(
            account_name,
            derived_list,
            conditional_ratio_threshold,
            top_n,
            allowed_node_names=allowed,
            node_belonging_dim=node_belonging_dim if node_belonging_dim else None,
        )

        # 批量匹配所有节点与帖子选题点
        if items and post_id:
            node_names = [x["节点名称"] for x in items]
            matched_results = await match_derivation_to_post_points(node_names, account_name, post_id)
            node_match_map: dict[str, list] = {}
            for m in matched_results:
                node_match_map.setdefault(m["推导选题点"], []).append({
                    "帖子选题点": m["帖子选题点"],
                    "匹配分数": m["匹配分数"],
                })
            for item in items:
                matches = node_match_map.get(item["节点名称"], [])
                item["帖子选题点匹配"] = matches if matches else "无"

        # [临时] 仅保留有帖子选题点匹配的记录（过滤掉「无」），方便后续删除
        items = [x for x in items if isinstance(x.get("帖子选题点匹配"), list)]

        if not items:
            output = f"未找到条件概率 >= {conditional_ratio_threshold} 的节点"
        else:
            lines = []
            for x in items:
                match_info = x.get("帖子选题点匹配", "无")
                if isinstance(match_info, list):
                    match_str = "、".join(f"{m['帖子选题点']}({m['匹配分数']})" for m in match_info)
                else:
                    match_str = str(match_info)
                dim_label = x.get("所属维度", "—")
                lines.append(
                    f"- {x['节点名称']}\t条件概率={x['条件概率']}\t父节点={x['父节点名称']}\t所属维度={dim_label}\t帖子选题点匹配={match_str}"
                )
            output = "\n".join(lines)
        return ToolResult(
            title=f"条件概率节点 ({account_name}, 阈值={conditional_ratio_threshold})",
            output=output,
            metadata={
                "account_name": account_name,
                "threshold": conditional_ratio_threshold,
                "top_n": top_n,
                "count": len(items),
                "round": int(round),
                "log_id": str(log_id).strip() if log_id else "",
                "dimension_filter": {
                    "derived_dim_nodes": derived_dim_names,
                    "allowed_descendant_count": len(allowed) if allowed is not None else None,
                    "source": dim_source or ("未提供 log_id，未按维度收窄" if not (log_id and str(log_id).strip()) else ""),
                },
            },
        )
    except Exception as e:
        return ToolResult(
            title="按条件概率查询节点失败",
            output=str(e),
            error=str(e),
        )


def main() -> None:
    """本地测试：用家有大志账号测常量节点与条件概率节点，有 agent 时再跑一遍 tool 接口。"""
    import asyncio

    account_name = "家有大志"
    post_id = "68fb6a5c000000000302e5de"
    # derived_items = [
    #     {"topic": "分享", "source_node": "分享"},
    #     {"topic": "叙事结构", "source_node": "叙事结构"},
    # ]
    derived_items = [{"topic":"分享","source_node":"分享"},{"topic":"叙事结构","source_node":"叙事编排"},{"topic":"幽默化标题","source_node":"幽默化标题"},{"source_node":"叙事结构","topic":"叙事结构"},{"topic":"夸张堆叠","source_node":"夸张转化"},{"topic":"居家生活场景","source_node":"生活场景"},{"topic":"图片文字","source_node":"图片文字"},{"source_node":"补充说明式","topic":"补充说明式"},{"topic":"标题","source_node":"标题"},{"topic":"递进式","source_node":"递进式"},{"source_node":"荒诞夸张","topic":"夸张堆叠"},{"topic":"图片文字","source_node":"图文组合"},{"topic":"补充说明式","source_node":"解读说明"},{"source_node":"分步骤","topic":"递进式"},{"source_node":"视觉证据","topic":"拖鞋物证"},{"topic":"鞋架","source_node":"家居器具"},{"topic":"柴犬形象","source_node":"形象演绎"},{"topic":"叙事结构","source_node":"结构编排"},{"topic":"图片文字","source_node":"图文编排"},{"source_node":"形象","topic":"柴犬形象"},{"source_node":"版面结构","topic":"叙事结构"},{"source_node":"夸张变形","topic":"夸张堆叠"},{"source_node":"夸张造型","topic":"夸张堆叠"},{"topic":"夸张堆叠","source_node":"夸张穿戴法"}]
    conditional_ratio_threshold = 0.2
    top_n = 2000

    # # 1）常量节点（核心函数，无匹配）
    # constant_nodes = get_constant_nodes(account_name)
    # print(f"账号: {account_name} — 常量节点共 {len(constant_nodes)} 个（前 50 个）:")
    # for x in constant_nodes[:50]:
    #     print(f"  - {x['节点名称']}\t概率={x['概率']}\t{x['常量类型']}")
    # print()
    #
    # # 2）条件概率节点（核心函数）
    # derived_list = _parse_derived_list(derived_items)
    # ratio_nodes = get_nodes_by_conditional_ratio(
    #     account_name, derived_list, conditional_ratio_threshold, top_n
    # )
    # print(f"条件概率节点 阈值={conditional_ratio_threshold}, top_n={top_n}, 共 {len(ratio_nodes)} 个:")
    # for x in ratio_nodes:
    #     print(f"  - {x['节点名称']}\t条件概率={x['条件概率']}\t父节点={x['父节点名称']}")
    # print()

    # 3）有 agent 时通过 tool 接口再跑一遍（含帖子选题点匹配）
    if ToolResult is not None:
        async def run_tools():
            # r1 = await find_tree_constant_nodes(account_name, post_id=post_id)
            # print("--- find_tree_constant_nodes ---")
            # print(r1.output[:2000] + "..." if len(r1.output) > 2000 else r1.output)
            r2 = await find_tree_nodes_by_conditional_ratio(
                account_name,
                post_id=post_id,
                derived_items=derived_items,
                conditional_ratio_threshold=conditional_ratio_threshold,
                top_n=top_n,
                round=6,
                log_id="20260318172724",
            )
            print("\n--- find_tree_nodes_by_conditional_ratio ---")
            print(r2.output)
        asyncio.run(run_tools())


if __name__ == "__main__":
    main()