import pandas as pd
import numpy as np
from pathlib import Path

# CSV 로드
df = pd.read_csv("results.csv")

# QC용 중복 영상 ID
qc_video = "Wan2.2/Wan2.2_BodyWeightSquats_03_g09_c02.mp4"

# 숫자형 변환
for col in ["action_consistency", "physical_plausibility"]:
    df[col] = pd.to_numeric(df[col], errors="coerce")

# QC용 영상만 필터링
dup_df = df[df["video_id"] == qc_video].copy()

qc_rows = []
for pid, g in dup_df.groupby("participant_id"):
    if len(g) < 2:
        continue  # 1회만 응답한 참가자 제외
    
    ac = g["action_consistency"].dropna().to_numpy()
    pp = g["physical_plausibility"].dropna().to_numpy()
    
    qc_rows.append({
        "participant_id": pid,
        "n_repeat": len(g),
        "ac_mean": np.mean(ac),
        "ac_std": np.std(ac, ddof=0),
        "ac_range": np.max(ac) - np.min(ac),
        "ac_mae": np.mean(np.abs(ac - np.mean(ac))),
        "pp_mean": np.mean(pp),
        "pp_std": np.std(pp, ddof=0),
        "pp_range": np.max(pp) - np.min(pp),
        "pp_mae": np.mean(np.abs(pp - np.mean(pp))),
    })

qc = pd.DataFrame(qc_rows)

# 품질 판정 (임계값은 필요 시 조정)
qc["low_quality_flag"] = (
    (qc["ac_std"] > 2.0) | (qc["ac_range"] > 4.0) |
    (qc["pp_std"] > 2.0) | (qc["pp_range"] > 4.0)
)

# 저장
Path("qc_outputs").mkdir(exist_ok=True)
qc_path = "qc_outputs/participant_qc_single_duplicate.csv"
qc.to_csv(qc_path, index=False)
print(f"Saved to {qc_path}")

# 결과 미리보기
print(qc.sort_values("low_quality_flag", ascending=False))