perfectionist

notebooks/gemini_rer.py

@@ -0,0 +1,249 @@
+import pandas as pd
+
+
+def mifflin_st_jeor(weight_kg, height_cm, age_years, sex):
+    """
+    Compute predicted RMR with Mifflin St Jeor.
+    sex: 'male' or 'female'
+    """
+    base = 10.0 * weight_kg + 6.25 * height_cm - 5.0 * age_years
+    if sex.lower().startswith("m"):
+        return base + 5.0
+    else:
+        return base - 161.0
+
+
+def classify_metabolism(measured_kcal_day, predicted_kcal_day):
+    """
+    Classify metabolic rate relative to prediction.
+    Returns (label, ratio).
+    """
+    ratio = measured_kcal_day / predicted_kcal_day
+
+    if ratio < 0.70:
+        label = "very slow"
+    elif ratio < 0.90:
+        label = "slow"
+    elif ratio <= 1.10:
+        label = "average"
+    elif ratio <= 1.30:
+        label = "fast"
+    else:
+        label = "very fast"
+
+    return label, ratio
+
+
+def find_sampling_window(df):
+    """
+    Derive number of samples that represent about 2 minutes.
+    """
+    dt = df["T(sec)"].diff().median()
+    if dt is None or dt <= 0:
+        raise ValueError("Invalid time step in T(sec)")
+
+    samples = int(round(120.0 / dt))
+    if samples < 1:
+        samples = 1
+    return samples
+
+
+def rolling_stable_window(df, window_samples):
+    """
+    Find the most stable 2-minute window using rolling standard deviation.
+    Returns:
+        means_series, t_start, t_end
+    """
+    cols_mean = [
+        "VO2(ml/min)",
+        "VCO2(ml/min)",
+        "VE(l/min)",
+        "VT(l)",
+        "BF(bpm)",
+        "EE(kcal/min)",
+        "RER",
+        "CARBS(%)",
+        "FAT(%)",
+    ]
+
+    cols_std = [
+        "VO2(ml/min)",
+        "VCO2(ml/min)",
+        "VE(l/min)",
+        "VT(l)",
+        "BF(bpm)",
+    ]
+
+    roll_mean = df[cols_mean].rolling(window_samples, min_periods=window_samples).mean()
+    roll_std = df[cols_std].rolling(window_samples, min_periods=window_samples).std()
+
+    # Sum std devs to get stability score; use skipna=False to preserve NaN for incomplete windows
+    stability_score = roll_std.sum(axis=1, skipna=False)
+
+    # Find index with lowest stability score (dropna to ignore incomplete windows)
+    best_idx = stability_score.dropna().idxmin()
+
+    means_series = roll_mean.loc[best_idx].copy()
+
+    start_idx = max(best_idx - window_samples + 1, 0)
+    end_idx = best_idx
+
+    t_start = float(df["T(sec)"].iloc[start_idx])
+    t_end = float(df["T(sec)"].iloc[end_idx])
+
+    return means_series, t_start, t_end
+
+
+def manual_window_means(df, t_start, t_end):
+    """
+    Compute mean values inside a user-selected time window.
+    """
+    mask = (df["T(sec)"] >= t_start) & (df["T(sec)"] <= t_end)
+    slice_df = df.loc[mask].copy()
+
+    if slice_df.empty:
+        raise ValueError("Manual window has no rows inside T(sec) range")
+
+    cols = [
+        "VO2(ml/min)",
+        "VCO2(ml/min)",
+        "VE(l/min)",
+        "VT(l)",
+        "BF(bpm)",
+        "EE(kcal/min)",
+        "RER",
+        "CARBS(%)",
+        "FAT(%)",
+    ]
+
+    means = slice_df[cols].mean()
+    return means, float(t_start), float(t_end)
+
+
+def load_pnoe_csv(path):
+    """
+    Load and clean a PNOE CSV file.
+    """
+    df = pd.read_csv(path, sep=";")
+
+    numeric_cols = [
+        "T(sec)",
+        "VO2(ml/min)",
+        "VCO2(ml/min)",
+        "RER",
+        "VE(l/min)",
+        "VT(l)",
+        "BF(bpm)",
+        "EE(kcal/min)",
+        "CARBS(%)",
+        "FAT(%)",
+    ]
+
+    for col in numeric_cols:
+        df[col] = pd.to_numeric(df[col], errors="coerce")
+
+    df = df.dropna(subset=["VO2(ml/min)", "EE(kcal/min)"]).reset_index(drop=True)
+    return df
+
+
+def analyze_pnoe_rmr(
+    path,
+    weight_kg,
+    height_cm,
+    age_years,
+    sex,
+    subject_name=None,
+    test_date=None,
+    manual_window=None,
+):
+    """
+    Analyze resting RMR from a PNOE CSV file.
+
+    manual_window:
+        None for automatic stable window
+        or (t_start_sec, t_end_sec) for user-chosen window
+    """
+    df = load_pnoe_csv(path)
+    window_samples = find_sampling_window(df)
+
+    # Automatic stable window
+    auto_means, auto_t_start, auto_t_end = rolling_stable_window(df, window_samples)
+
+    # Manual override if provided
+    manual_means = None
+    manual_t_start = None
+    manual_t_end = None
+
+    if manual_window is not None:
+        t_start_manual, t_end_manual = manual_window
+        manual_means, manual_t_start, manual_t_end = manual_window_means(
+            df, t_start_manual, t_end_manual
+        )
+        chosen_source = "manual"
+        chosen_means = manual_means
+        chosen_t_start = manual_t_start
+        chosen_t_end = manual_t_end
+    else:
+        chosen_source = "auto"
+        chosen_means = auto_means
+        chosen_t_start = auto_t_start
+        chosen_t_end = auto_t_end
+
+    kcal_per_min = float(chosen_means["EE(kcal/min)"])
+    rmr_kcal_day = kcal_per_min * 1440.0
+
+    predicted_kcal_day = mifflin_st_jeor(weight_kg, height_cm, age_years, sex)
+    label, ratio = classify_metabolism(rmr_kcal_day, predicted_kcal_day)
+
+    def pack_metrics(prefix, means, t_start, t_end):
+        if means is None:
+            return {}
+        return {
+            f"{prefix}_window_start_sec": t_start,
+            f"{prefix}_window_end_sec": t_end,
+            f"{prefix}_VO2_L_min": float(means["VO2(ml/min)"]) / 1000.0,
+            f"{prefix}_VCO2_L_min": float(means["VCO2(ml/min)"]) / 1000.0,
+            f"{prefix}_VE_L_min": float(means["VE(l/min)"]),
+            f"{prefix}_VT_L": float(means["VT(l)"]),
+            f"{prefix}_BF_bpm": float(means["BF(bpm)"]),
+            f"{prefix}_RER": float(means["RER"]),
+            f"{prefix}_Fat_percent": float(means["FAT(%)"]),
+            f"{prefix}_Carb_percent": float(means["CARBS(%)"]),
+            f"{prefix}_kcal_per_min": float(means["EE(kcal/min)"]),
+        }
+
+    result = {
+        "subject_name": subject_name,
+        "test_date": test_date,
+        "sex": sex,
+        "weight_kg": weight_kg,
+        "height_cm": height_cm,
+        "age_years": age_years,
+        "chosen_window_source": chosen_source,
+        "chosen_window_start_sec": chosen_t_start,
+        "chosen_window_end_sec": chosen_t_end,
+        "RMR_kcal_day": rmr_kcal_day,
+        "Mifflin_kcal_day": predicted_kcal_day,
+        "Measured_to_Mifflin_ratio": ratio,
+        "Metabolic_classification": label,
+    }
+
+    result.update(pack_metrics("auto", auto_means, auto_t_start, auto_t_end))
+    result.update(pack_metrics("manual", manual_means, manual_t_start, manual_t_end))
+
+    return result
+
+
+result = analyze_pnoe_rmr(
+    path="/home/oluwasanmi/Documents/Work/MKD/report_generation/data/Pnoe_20250729_1550-Moran_Keirstyn.csv",
+    weight_kg=56,
+    height_cm=162,
+    age_years=34,
+    sex="female",
+    subject_name="Cullen Pacas",
+    test_date="2025-11-12",
+    manual_window=None,  # or (t_start_sec, t_end_sec)
+)
+
+for key, value in result.items():
+    print(f"{key}: {value}")