rows, cols = df.shape
mem_bytes = int(df.memory_usage(deep=True).sum())
mem_mb = round(mem_bytes / (1024*1024), 2)

# Build Integrity Notes lists and full flagged table
flag_rows = []
for c in df.columns:
    f = []
    if c in id_like: f.append('id-like')
    if c in hi_card: f.append('high-card')
    if c in consts:  f.append('constant')
    if f:
        flag_rows.append({
           'Column': c,
           'Flags': ', '.join(f),
           'Type': dtypes[c],
           'Unique Values': int(uniques.get(c,0)),
           'Unique Ratio': float(uniq_ratio.get(c,0)),
           'Missing (%)': float(miss_pct.get(c,0))
        })
flagged_df = pd.DataFrame(flag_rows)
flagged_df

df[['year', 'make', 'model', 'trim', 'body', 'transmission', 'vin', 'state', 'condition', 'odometer', 'color', 'interior', 'seller', 'mmr', 'sellingprice', 'saledate']].head(8)

df[['year', 'make', 'model', 'trim', 'body', 'transmission', 'vin', 'state', 'condition', 'odometer', 'color', 'interior', 'seller', 'mmr', 'sellingprice', 'saledate']].describe()

dtypes = df.dtypes.astype(str)
nonnull = df.notnull().sum()
miss_pct = (df.isnull().mean()*100).round(1)
uniques = df.nunique(dropna=True)
n = max(len(df), 1)
uniq_ratio = (uniques / n).fillna(0.0)
def _sample_vals(s, k=3):
    vals = pd.unique(s.dropna().astype(str))[:k]
    return ', '.join(map(str, vals)) if len(vals) else ''
flags_col = []
for c in df.columns:
    flags=[]
    if uniques.get(c,0) <= 1: flags.append('constant')
    if uniq_ratio.get(c,0) >= 0.95 and 'datetime' not in dtypes[c].lower(): flags.append('id-like')
    if dtypes[c].startswith('object') and uniq_ratio.get(c,0) > 0.5 and 'id-like' not in flags: flags.append('high-card')
    flags_col.append(', '.join(flags))
profile_df = pd.DataFrame({
  'Column': df.columns,
  'Type': dtypes.values,
  'Non-Null Count': nonnull.values,
  'Missing (%)': miss_pct.values,
  'Unique Values': uniques.values,
  'Sample Values': [ _sample_vals(df[c]) for c in df.columns ],
  'Flags': flags_col
})
profile_df

pd.DataFrame({
    'Column': df.columns,
    'Type': df.dtypes.astype(str).values,
    'Non-Null Count': df.notnull().sum().values,
    'Unique Values': df.nunique().values
})

num_cols = df.select_dtypes(include='number').columns.tolist()[:max_cols]
df_num = df[num_cols]
df_num = df_num.sample(max_rows, random_state=0) if len(df_num) > max_rows else df_num
med = df_num.median(); mad = (df_num - med).abs().median()
rz = 0.6745 * (df_num - med) / mad.replace(0, np.nan)
row_score = rz.abs().max(axis=1)
top3 = row_score.nlargest(3)

dfn = df.select_dtypes(include='number').iloc[:, :max_cols]
dfn = dfn.sample(max_rows, random_state=0) if len(dfn) > max_rows else dfn
# rank columns by Tukey outliers (1.5*IQR) and plot violins with inner box

nulls = df.isnull().sum()
nulls_pct = (df.isnull().mean() * 100).round(1)
missing_df = pd.DataFrame({
    'Column': df.columns,
    'Missing Values': nulls.values,
    'Missing (%)': nulls_pct.values
})\nmissing_df[missing_df['Missing Values'] > 0]

nulls = df.isnull().sum();
nulls_pct = (
    df.isnull().mean()*100
).round(1)
missing_df = pd.DataFrame({
    'Column': df.columns,
    'Missing Values': nulls.values,
    'Missing (%)': nulls_pct.values
})

top_miss = (
    missing_df[missing_df['Missing Values'] > 0]
       .sort_values('Missing (%)', ascending=False)
       .loc[:, ['Column', 'Missing (%)']]
       .head(20)
       .reset_index(drop=True)
)
top_miss

dist_col = '<chosen categorical>'
s = df[dist_col].astype('object').where(~df[dist_col].isna(), other='Missing')
vc = s.value_counts(dropna=False)
top_k = 8  # Top-8 + Other (+ Missing)