test_tasks/Medical-Data-Standardization/hepatic-panel-harmonization/solution/tools/harmonize.py

import pandas as pd

def parse_value(value):
    if pd.isna(value):
        return None
    s = str(value).strip()
    if s == '' or s.lower() == 'nan':
        return None
    if ',' in s:
        s = s.replace(',', '.')
    return float(s)

def normalize_frame(df, specs, id_column=None):
    numeric_cols = [c for c in df.columns if c != id_column]
    missing_mask = df[numeric_cols].applymap(lambda x: pd.isna(x) or str(x).strip() == '' or str(x).strip().lower() == 'nan').any(axis=1)
    df = df.loc[~missing_mask].copy()
    for col in numeric_cols:
        mode, factor, lo, hi = specs[col]
        df[col] = df[col].apply(parse_value)
        def convert(v):
            if v is None:
                return None
            if lo <= v <= hi:
                return v
            if mode == 'single':
                c = v / factor
                return c if lo <= c <= hi else v
            if mode == 'single-reverse':
                c = v * factor
                return c if lo <= c <= hi else v
            return v
        df[col] = df[col].apply(convert)
        df[col] = df[col].apply(lambda x: f"{x:.2f}")
    return df