monitor.py

import numpy as np
import pandas as pd
from typing import Optional, Dict, Any


def _safe_change(current, base):
    """
    増減率 current / base - 1
    base=0 の場合は NaN
    """
    if pd.isna(current) or pd.isna(base) or base == 0:
        return np.nan

    return current / base - 1.0


def _robust_z(current, history):
    """
    Median / MAD を使った Robust Z-score
    """
    x = (
        pd.Series(history, dtype="float64")
        .replace([np.inf, -np.inf], np.nan)
        .dropna()
    )

    if len(x) == 0 or pd.isna(current):
        return np.nan

    median = x.median()
    mad = (x - median).abs().median()

    if mad == 0:
        if current == median:
            return 0.0

        return np.inf if current > median else -np.inf

    return (current - median) / (1.4826 * mad)


def _normalize_month(series):
    """
    以下のような年月表現を Period[M] に統一

    202608
    "202608"
    "2026-08"
    datetime
    """

    if isinstance(series.dtype, pd.PeriodDtype):
        return series.dt.asfreq("M")

    if pd.api.types.is_datetime64_any_dtype(series):
        return series.dt.to_period("M")

    text = series.astype(str).str.strip()

    is_yyyymm = text.str.fullmatch(r"\d{6}")

    out = pd.Series(
        pd.NaT,
        index=series.index,
        dtype="datetime64[ns]"
    )

    if is_yyyymm.any():
        out.loc[is_yyyymm] = pd.to_datetime(
            text.loc[is_yyyymm],
            format="%Y%m",
            errors="coerce"
        )

    if (~is_yyyymm).any():
        out.loc[~is_yyyymm] = pd.to_datetime(
            text.loc[~is_yyyymm],
            errors="coerce"
        )

    return out.dt.to_period("M")


def monitor_trade(
    df: pd.DataFrame,
    *,
    month_col: str,
    country_col: str,
    value_col: str,
    quantity_col: str,
    target_month: Optional[str] = None,
    quantity_unit_col: Optional[str] = None,
    top_n: int = 10,
    event_min_share: float = 0.0,
    event_min_value: float = 0.0,
) -> Dict[str, Any]:
    """
    貿易統計モニタリング用集計関数

    想定:
        対象HSについて、
        分析対象月 + 過去12か月
        = 合計13か月分の輸入申告データを渡す。

    Parameters
    ----------
    df:
        輸入申告DataFrame

    month_col:
        年月カラム
        例: "年月"

    country_col:
        原産国カラム
        例: "原産国"

    value_col:
        取引額カラム
        例: "課税価格"

    quantity_col:
        統計数量カラム
        例: "統計数量"

    target_month:
        分析対象月
        例: "2026-08"

        Noneの場合はdf内の最新月

    quantity_unit_col:
        統計数量単位
        例: "統計数量単位"

        指定した場合、複数単位が混在していればエラーにする。

    top_n:
        国別ランキング表示件数

    event_min_share:
        新規国・消失国として抽出する最低シェア
        例:
            0.01 = 1%

    event_min_value:
        新規国・消失国として抽出する最低金額


    Returns
    -------
    dict

    {
        "target_month":
            分析対象月

        "yoy_month":
            前年同月

        "monthly":
            過去13か月の月次推移

        "metric_summary":
            金額・数量・単価の
            YoY / MoM / 3か月平均比 /
            12か月中央値比 / Robust Z

        "concentration":
            CR1 / CR3 / CR5 / HHI /
            有効供給国数 / 構造変化指数

        "country_all":
            全原産国比較

        "country_top_value":
            金額TOP N

        "country_top_quantity":
            数量TOP N

        "events":
            新規国・消失国

        "warnings":
            警告
    }
    """

    # -------------------------------------------------
    # カラム確認
    # -------------------------------------------------

    required_cols = [
        month_col,
        country_col,
        value_col,
        quantity_col,
    ]

    if quantity_unit_col is not None:
        required_cols.append(quantity_unit_col)

    missing_cols = [
        c for c in required_cols
        if c not in df.columns
    ]

    if missing_cols:
        raise KeyError(
            f"存在しないカラムがあります: {missing_cols}"
        )

    # -------------------------------------------------
    # 必要列のみコピー
    # -------------------------------------------------

    x = df[required_cols].copy()

    x["_month"] = _normalize_month(
        x[month_col]
    )

    x["_value"] = pd.to_numeric(
        x[value_col],
        errors="coerce"
    )

    x["_quantity"] = pd.to_numeric(
        x[quantity_col],
        errors="coerce"
    )

    x["_country"] = (
        x[country_col]
        .astype("string")
        .fillna("(不明)")
    )

    x = x.dropna(
        subset=[
            "_month",
            "_value",
            "_quantity"
        ]
    )

    if x.empty:
        raise ValueError(
            "有効なデータがありません。"
            "年月・金額・数量カラムを確認してください。"
        )

    # -------------------------------------------------
    # 分析対象月
    # -------------------------------------------------

    if target_month is None:
        target = x["_month"].max()
    else:
        target = pd.Period(
            target_month,
            freq="M"
        )

    start_month = target - 12
    yoy_month = target - 12
    prev_month = target - 1

    # 13か月分に限定
    x = x[
        x["_month"].between(
            start_month,
            target
        )
    ].copy()

    if x.empty:
        raise ValueError(
            f"{start_month}〜{target} の"
            "データがありません。"
        )

    # -------------------------------------------------
    # 統計数量単位チェック
    # -------------------------------------------------

    if quantity_unit_col is not None:

        units = (
            x[quantity_unit_col]
            .dropna()
            .astype(str)
            .unique()
        )

        if len(units) > 1:
            raise ValueError(
                "統計数量の単位が複数あります。\n"
                f"{list(units)}\n"
                "単位ごと、またはHSごとに分けて"
                "集計してください。"
            )

    # -------------------------------------------------
    # 13か月存在確認
    # -------------------------------------------------

    expected_months = pd.period_range(
        start=start_month,
        end=target,
        freq="M"
    )

    months_present = set(
        x["_month"].unique()
    )

    missing_months = [
        str(m)
        for m in expected_months
        if m not in months_present
    ]

    warnings = []

    if missing_months:
        warnings.append(
            "対象HSのレコードがない月を"
            "0として扱いました: "
            + ", ".join(missing_months)
        )

    # =================================================
    # 1. 月次全体集計
    # =================================================

    monthly = (
        x
        .groupby(
            "_month",
            observed=True
        )
        .agg(
            value=(
                "_value",
                "sum"
            ),
            quantity=(
                "_quantity",
                "sum"
            ),
        )
        .reindex(
            expected_months,
            fill_value=0.0
        )
    )

    monthly.index.name = "month"

    monthly["unit_price"] = (
        monthly["value"]
        .div(
            monthly["quantity"]
            .replace(0, np.nan)
        )
    )

    # =================================================
    # 2. 全体指標
    # =================================================

    summary_rows = []

    for metric in [
        "value",
        "quantity",
        "unit_price",
    ]:

        s = monthly[metric]

        current = s.loc[target]

        yoy_base = s.loc[yoy_month]

        mom_base = s.loc[prev_month]

        history_12m = s.loc[
            start_month:prev_month
        ]

        previous_3m = s.loc[
            target - 3:prev_month
        ]

        summary_rows.append(
            {
                "metric": metric,

                "current":
                    current,

                "yoy_base":
                    yoy_base,

                # 前年同月比
                "yoy_pct":
                    _safe_change(
                        current,
                        yoy_base
                    ),

                # 前月比
                "mom_pct":
                    _safe_change(
                        current,
                        mom_base
                    ),

                # 直近3か月平均比
                "vs_prev3m_avg_pct":
                    _safe_change(
                        current,
                        previous_3m.mean()
                    ),

                # 過去12か月中央値比
                "vs_prev12m_median_pct":
                    _safe_change(
                        current,
                        history_12m.median()
                    ),

                # Robust Z
                "robust_z_12m":
                    _robust_z(
                        current,
                        history_12m
                    ),
            }
        )

    metric_summary = (
        pd.DataFrame(summary_rows)
        .set_index("metric")
    )

    # =================================================
    # 3. 国別集計
    # =================================================

    grouped = (
        x
        .groupby(
            [
                "_month",
                "_country",
            ],
            observed=True,
        )
        .agg(
            value=(
                "_value",
                "sum"
            ),
            quantity=(
                "_quantity",
                "sum"
            ),
        )
        .reset_index()
    )

    # 対象月
    current_country = (
        grouped[
            grouped["_month"] == target
        ][
            [
                "_country",
                "value",
                "quantity",
            ]
        ]
        .rename(
            columns={
                "value":
                    "value_current",

                "quantity":
                    "quantity_current",
            }
        )
    )

    # 前年同月
    previous_country = (
        grouped[
            grouped["_month"] == yoy_month
        ][
            [
                "_country",
                "value",
                "quantity",
            ]
        ]
        .rename(
            columns={
                "value":
                    "value_yoy_base",

                "quantity":
                    "quantity_yoy_base",
            }
        )
    )

    # 外部結合
    country = (
        current_country
        .merge(
            previous_country,
            on="_country",
            how="outer"
        )
        .fillna(
            {
                "value_current": 0.0,
                "quantity_current": 0.0,
                "value_yoy_base": 0.0,
                "quantity_yoy_base": 0.0,
            }
        )
    )

    # =================================================
    # 4. 国別 金額 / 数量 共通指標
    # =================================================

    for prefix in [
        "value",
        "quantity",
    ]:

        current = country[
            f"{prefix}_current"
        ]

        previous = country[
            f"{prefix}_yoy_base"
        ]

        current_total = current.sum()
        previous_total = previous.sum()

        # 差分
        country[
            f"{prefix}_delta"
        ] = (
            current
            - previous
        )

        # YoY
        country[
            f"{prefix}_yoy_pct"
        ] = np.where(
            previous != 0,
            current / previous - 1,
            np.nan
        )

        # シェア
        if current_total > 0:

            country[
                f"{prefix}_share_current"
            ] = (
                current
                / current_total
            )

        else:

            country[
                f"{prefix}_share_current"
            ] = 0.0

        if previous_total > 0:

            country[
                f"{prefix}_share_yoy_base"
            ] = (
                previous
                / previous_total
            )

        else:

            country[
                f"{prefix}_share_yoy_base"
            ] = 0.0

        # シェア前年差
        country[
            f"{prefix}_share_delta_pp"
        ] = (
            country[
                f"{prefix}_share_current"
            ]
            -
            country[
                f"{prefix}_share_yoy_base"
            ]
        ) * 100

        # 順位
        country[
            f"{prefix}_rank_current"
        ] = (
            current
            .where(current > 0)
            .rank(
                method="min",
                ascending=False
            )
        )

        country[
            f"{prefix}_rank_yoy_base"
        ] = (
            previous
            .where(previous > 0)
            .rank(
                method="min",
                ascending=False
            )
        )

        # 正なら順位上昇
        country[
            f"{prefix}_rank_change"
        ] = (
            country[
                f"{prefix}_rank_yoy_base"
            ]
            -
            country[
                f"{prefix}_rank_current"
            ]
        )

    # =================================================
    # 5. 国別単価
    # =================================================

    country["unit_price_current"] = (
        country["value_current"]
        .div(
            country[
                "quantity_current"
            ]
            .replace(
                0,
                np.nan
            )
        )
    )

    country["unit_price_yoy_base"] = (
        country["value_yoy_base"]
        .div(
            country[
                "quantity_yoy_base"
            ]
            .replace(
                0,
                np.nan
            )
        )
    )

    country["unit_price_yoy_pct"] = (
        country[
            "unit_price_current"
        ]
        .div(
            country[
                "unit_price_yoy_base"
            ]
        )
        .sub(1)
    )

    # =================================================
    # 6. 新規 / 消失国
    # =================================================

    country["status"] = np.select(

        [
            (
                country["value_current"] > 0
            )
            &
            (
                country["value_yoy_base"] == 0
            ),

            (
                country["value_current"] == 0
            )
            &
            (
                country["value_yoy_base"] > 0
            ),
        ],

        [
            "new",
            "disappeared",
        ],

        default="continuing"
    )

    # 新規・消失国のうち
    # ある程度重要なものだけ抽出可能
    country["event_flag"] = (

        (
            country["status"]
            != "continuing"
        )

        &

        (
            (
                country[
                    [
                        "value_share_current",
                        "value_share_yoy_base",
                    ]
                ]
                .max(axis=1)
                >= event_min_share
            )

            |

            (
                country[
                    [
                        "value_current",
                        "value_yoy_base",
                    ]
                ]
                .max(axis=1)
                >= event_min_value
            )
        )
    )

    # =================================================
    # 7. 集中度
    # =================================================

    def concentration(prefix):

        current_share = country[
            f"{prefix}_share_current"
        ]

        previous_share = country[
            f"{prefix}_share_yoy_base"
        ]

        current_values = country[
            f"{prefix}_current"
        ]

        previous_values = country[
            f"{prefix}_yoy_base"
        ]

        current_total = (
            current_values.sum()
        )

        previous_total = (
            previous_values.sum()
        )

        def calc_one(
            shares,
            values,
            total,
        ):

            if total <= 0:
                return {
                    "cr1": np.nan,
                    "cr3": np.nan,
                    "cr5": np.nan,
                    "hhi": np.nan,
                    "effective_suppliers":
                        np.nan,
                    "supplier_count": 0,
                }

            hhi = (
                shares ** 2
            ).sum()

            return {
                "cr1":
                    shares
                    .nlargest(1)
                    .sum(),

                "cr3":
                    shares
                    .nlargest(3)
                    .sum(),

                "cr5":
                    shares
                    .nlargest(5)
                    .sum(),

                "hhi":
                    hhi,

                "effective_suppliers":
                    (
                        1 / hhi
                        if hhi > 0
                        else np.nan
                    ),

                "supplier_count":
                    int(
                        (
                            values > 0
                        ).sum()
                    ),
            }

        now = calc_one(
            current_share,
            current_values,
            current_total,
        )

        before = calc_one(
            previous_share,
            previous_values,
            previous_total,
        )

        # Total Variation Distance
        # 0 = 構造変化なし
        # 1 = 完全に別の供給構造
        if (
            current_total > 0
            and previous_total > 0
        ):

            structure_change = (
                0.5
                *
                (
                    current_share
                    - previous_share
                )
                .abs()
                .sum()
            )

        else:
            structure_change = np.nan

        return {
            "basis":
                prefix,

            "cr1_current":
                now["cr1"],

            "cr1_yoy_base":
                before["cr1"],

            "cr1_delta_pp":
                (
                    (
                        now["cr1"]
                        - before["cr1"]
                    )
                    * 100
                    if (
                        pd.notna(
                            now["cr1"]
                        )
                        and
                        pd.notna(
                            before["cr1"]
                        )
                    )
                    else np.nan
                ),

            "cr3_current":
                now["cr3"],

            "cr3_yoy_base":
                before["cr3"],

            "cr3_delta_pp":
                (
                    (
                        now["cr3"]
                        - before["cr3"]
                    )
                    * 100
                    if (
                        pd.notna(
                            now["cr3"]
                        )
                        and
                        pd.notna(
                            before["cr3"]
                        )
                    )
                    else np.nan
                ),

            "cr5_current":
                now["cr5"],

            "cr5_yoy_base":
                before["cr5"],

            "cr5_delta_pp":
                (
                    (
                        now["cr5"]
                        - before["cr5"]
                    )
                    * 100
                    if (
                        pd.notna(
                            now["cr5"]
                        )
                        and
                        pd.notna(
                            before["cr5"]
                        )
                    )
                    else np.nan
                ),

            "hhi_current":
                now["hhi"],

            "hhi_yoy_base":
                before["hhi"],

            "hhi_delta":
                (
                    now["hhi"]
                    - before["hhi"]
                    if (
                        pd.notna(
                            now["hhi"]
                        )
                        and
                        pd.notna(
                            before["hhi"]
                        )
                    )
                    else np.nan
                ),

            "effective_suppliers_current":
                now[
                    "effective_suppliers"
                ],

            "effective_suppliers_yoy_base":
                before[
                    "effective_suppliers"
                ],

            "supplier_count_current":
                now[
                    "supplier_count"
                ],

            "supplier_count_yoy_base":
                before[
                    "supplier_count"
                ],

            "structure_change_index":
                structure_change,
        }

    concentration_df = (
        pd.DataFrame(
            [
                concentration("value"),
                concentration("quantity"),
            ]
        )
        .set_index("basis")
    )

    # 元の国カラム名に戻す
    country = country.rename(
        columns={
            "_country":
                country_col
        }
    )

    # =================================================
    # 8. TOP N
    # =================================================

    country_top_value = (
        country
        .sort_values(
            [
                "value_current",
                "value_yoy_base",
            ],
            ascending=False
        )
        .head(top_n)
        .reset_index(drop=True)
    )

    country_top_quantity = (
        country
        .sort_values(
            [
                "quantity_current",
                "quantity_yoy_base",
            ],
            ascending=False
        )
        .head(top_n)
        .reset_index(drop=True)
    )

    # =================================================
    # 9. 新規 / 消失国
    # =================================================

    events = (
        country[
            country["event_flag"]
        ]
        .sort_values(
            [
                "status",
                "value_current",
                "value_yoy_base",
            ],
            ascending=[
                True,
                False,
                False,
            ]
        )
        .reset_index(drop=True)
    )

    country_all = (
        country
        .sort_values(
            [
                "value_current",
                "value_yoy_base",
            ],
            ascending=False
        )
        .reset_index(drop=True)
    )

    # =================================================
    # Return
    # =================================================

    return {

        "target_month":
            str(target),

        "yoy_month":
            str(yoy_month),

        "monthly":
            monthly.reset_index(),

        "metric_summary":
            metric_summary,

        "concentration":
            concentration_df,

        "country_all":
            country_all,

        "country_top_value":
            country_top_value,

        "country_top_quantity":
            country_top_quantity,

        "events":
            events,

        "warnings":
            warnings,
    }