想做哪种统计？先帮我确认这些关键信息：

• 联赛与赛季范围（例如 2024-25 英超，或最近3赛季）
• 数据来源（你有现成CSV/Excel吗？还是让我抓 FBref/Understat/StatsBomb 开源数据）
• 指标口味与输出形式（榜单/表格/雷达图；导出 CSV/Excel/图片；是否分位置）
• 最低上场门槛（如 ≥900 分钟）与是否做每90分钟标准化与百分位

建议统计维度（按位置分层并做每90 + 百分位）：

• 进攻产出：非点球进球、助攻、npxG、xA、射门、关键传球、SCA/GCA
• 推进威胁：推进传球/带球、进攻三区传球、进禁区传球
• 终结效率：进球 - xG、每射门进球、射正率
• 组织与连接：传球尝试/成功率、三分之一/禁区传球成功率
• 防守与对抗：抢断、拦截、压迫、空中对抗胜率、地面对抗
• 可用性：出场分钟、首发次数、位置标签（FW/MF/DF，或更细分）
• 综合：按权重或PCA构建“关键球员评分”，并输出榜单

下面给你一个可直接跑的 Python/pandas 脚本雏形：读取一个球员赛季汇总 CSV，计算 per90、百分位与综合评分，并导出榜单。你只需把列名映射到你实际数据即可。

# file: key_players.py
import pandas as pd
import numpy as np

# 配置
INPUT = "players.csv"          # 你的球员赛季级数据
OUT_LEADERS = "leaders.csv"    # 联赛榜单
OUT_BY_TEAM = "by_team_top.csv"
MIN_MINUTES = 900              # 最低出场门槛
POS_MAP = {"F":"FW","M":"MF","D":"DF","GK":"GK"}  # 粗分位置

# 期望的列（可缺省，脚本会兜底为0）
EXPECTED = {
    "player":"player", "team":"team", "pos":"pos", "minutes":"minutes",
    "goals":"goals", "pens_made":"pens_made", "assists":"assists",
    "shots":"shots", "key_passes":"key_passes",
    "xG":"xG", "xA":"xA",
    "sca":"sca", "gca":"gca",
    "progressive_passes":"prog_passes",
    "progressive_carries":"prog_carries",
    "passes_into_final_third":"p_final3",
    "passes_into_penalty_area":"p_box",
    "tackles":"tackles", "interceptions":"interceptions",
    "pressures":"pressures",
    "aerials_won":"aerials_won", "aerials_lost":"aerials_lost"
}

def load():
    df = pd.read_csv(INPUT)
    # 标准化列名：尽量匹配 EXPECTED 指向的名字
    # 你可以直接把 EXPECTED 的 value 改成你CSV的真实列名
    rename = {v:k for k,v in EXPECTED.items() if v in df.columns}
    df = df.rename(columns=rename)
    # 为缺失列补0
    for k in EXPECTED.keys():
        if k not in df.columns:
            df[k] = 0
    # 位置粗分
    def map_pos(p):
        if isinstance(p,str) and p:
            c = p[0].upper()
            return POS_MAP.get(c, c)
        return "UNK"
    df["pos_grp"] = df["pos"].apply(map_pos)
    return df

def per90(df):
    m = df["minutes"].replace(0, np.nan)
    def p90(x): return x / m * 90
    df["npxG"] = df["xG"] - 0.0  # 若有点球xG列可减去
    df["npxG90"] = p90(df["npxG"])
    df["xA90"] = p90(df["xA"])
    df["G90"] = p90(df["goals"])
    df["A90"] = p90(df["assists"])
    df["Shots90"] = p90(df["shots"])
    df["KP90"] = p90(df["key_passes"])
    df["SCA90"] = p90(df["sca"])
    df["GCA90"] = p90(df["gca"])
    df["ProgAct90"] = p90(df["progressive_passes"] + df["progressive_carries"])
    df["Final3P90"] = p90(df["passes_into_final_third"])
    df["BoxP90"] = p90(df["passes_into_penalty_area"])
    df["DefAct90"] = p90(df["tackles"] + df["interceptions"])
    df["Press90"] = p90(df["pressures"])
    # 空战胜率
    air_total = df["aerials_won"] + df["aerials_lost"]
    df["AerialWin%"] = np.where(air_total>0, df["aerials_won"]/air_total, np.nan)
    # 进攻综合
    df["npxG+xA90"] = df["npxG90"] + df["xA90"]
    # 终结效率（简单版）
    df["G-xG"] = df["goals"] - df["xG"]
    return df

def percentiles(df, cols, by="pos_grp"):
    # 按位置分组做百分位
    def pct(s, v): 
        # 以nan安全的方式计算百分位
        return (s < v).mean() if np.isfinite(v) else np.nan
    for c in cols:
        pname = c + "_pct"
        df[pname] = np.nan
        for g, sub in df.groupby(by):
            s = sub[c].astype(float)
            ranks = s.rank(pct=True, method="average")
            df.loc[sub.index, pname] = ranks.values
    return df



def composite_score(df):
    # 不同位置可用不同权重，这里提供一个默认权重
    weights = {
        "FW": {"npxG90_pct":0.35, "xA90_pct":0.15, "Shots90_pct":0.10,
               "KP90_pct":0.10, "SCA90_pct":0.10, "ProgAct90_pct":0.05,
               "DefAct90_pct":0.05, "Press90_pct":0.05, "AerialWin%_pct":0.05},
        "MF": {"xA90_pct":0.20, "KP90_pct":0.15, "SCA90_pct":0.15,
               "ProgAct90_pct":0.15, "Final3P90_pct":0.10, "BoxP90_pct":0.05,
               "DefAct90_pct":0.10, "Press90_pct":0.05, "npxG90_pct":0.05},
        "DF": {"DefAct90_pct":0.30, "Press90_pct":0.10, "AerialWin%_pct":0.15,
               "ProgAct90_pct":0.10, "Final3P90_pct":0.10, "BoxP90_pct":0.05,
               "xA90_pct":0.05, "KP90_pct":0.05, "npxG90_pct":0.10},
        "GK": {}  # 本脚本不评GK
    }
    cols_for_pct = ["npxG90","xA90","Shots90","KP90","SCA90","GCA90",
                    "ProgAct90","Final3P90","BoxP90","DefAct90","Press90","AerialWin%"]
    df = percentiles(df, cols_for_pct + ["G-xG"], by="pos_grp")
    scores = []
    for idx, row in df.iterrows():
        w = weights.get(row["pos_grp"], weights["MF"])
        s = 0.0
        for k, alpha in w.items():
            v = row.get(k, np.nan)
            if np.isfinite(v):
                s += alpha * v
        scores.append(s)
    df["KeyScore"] = scores
    return df

def main():
    df = load()
    # 过滤分钟
    df = df[df["minutes"] >= MIN_MINUTES].copy()
    df = per90(df)
    df = composite_score(df)
    # 榜单
    leaders = (df.sort_values(["pos_grp","KeyScore"], ascending=[True, False])
                 .groupby("pos_grp")
                 .head(50)
                 .reset_index(drop=True))
    # 导出
    keep_cols = ["player","team","pos","pos_grp","minutes",
                 "npxG90","xA90","npxG+xA90","Shots90","KP90","SCA90","GCA90",
                 "ProgAct90","Final3P90","BoxP90","DefAct90","Press90","AerialWin%",
                 "G-xG","KeyScore"]
    leaders[keep_cols].to_csv(OUT_LEADERS, index=False)
    # 各队内前5
    by_team = (df.sort_values("KeyScore", ascending=False)
                 .groupby("team")
                 .head(5)
                 .reset_index(drop=True))
    by_team[["team","player","pos","KeyScore","npxG+xA90","ProgAct90","DefAct90"]].to_csv(OUT_BY_TEAM, index=False)
    print(f"Saved {OUT_LEADERS} and {OUT_BY_TEAM}")

if __name__ == "__main__":
    main()

使用方法

• 准备一个 players.csv，至少含这些列：player, team, pos, minutes, goals, assists, shots, key_passes, xG, xA；其余列缺失也可运行。
• 若你的列名不同，把脚本顶部 EXPECTED 右侧的值改成你实际列名。
• 运行：python key_players.py，会生成 leaders.csv（联赛榜单）与 by_team_top.csv（各队内Top5）。

需要我：

1. 直接对接你已有CSV做一次适配与输出图表；
2. 帮你抓取指定联赛赛季的数据并清洗成上述格式；
3. 调整权重/指标，做专属“关键球员评分”和雷达图。

给我一个样例数据头几行或说明目标联赛与赛季，我就继续完善到可复用的管道。