data_prep.py

# data_prep.py


# example: recency from last_purchase_days_ago (if exists)
if 'last_purchase_days_ago' in df.columns:
df['recency'] = df['last_purchase_days_ago']
else:
df['recency'] = np.nan


# example: tenure from signup_date
if 'signup_date' in df.columns:
df['signup_date'] = pd.to_datetime(df['signup_date'], errors='coerce')
df['tenure_days'] = (pd.Timestamp('today') - df['signup_date']).dt.days.fillna(df['signup_date'].median())
else:
df['tenure_days'] = np.nan


# keep numeric features and encode categorical later
return df




def prepare_features(df: pd.DataFrame, cat_cols=None, save_encoder_path=None):
df = df.copy()
if cat_cols is None:
cat_cols = [c for c in df.columns if df[c].dtype == 'object']


num_cols = [c for c in df.columns if pd.api.types.is_numeric_dtype(df[c])]


# One-hot encode categories (simple)
if len(cat_cols) > 0:
encoder = OneHotEncoder(handle_unknown='ignore', sparse=False)
cat_mat = encoder.fit_transform(df[cat_cols].astype(str))
cat_df = pd.DataFrame(cat_mat, columns=encoder.get_feature_names_out(cat_cols), index=df.index)
features = pd.concat([df[num_cols], cat_df], axis=1)
if save_encoder_path:
joblib.dump(encoder, save_encoder_path)
else:
features = df[num_cols]


return features.fillna(0)




if __name__ == '__main__':
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--input', default='data/customers_example.csv')
parser.add_argument('--out_features', default='data/features.parquet')
parser.add_argument('--save_encoder', default='data/ohe.joblib')
args = parser.parse_args()


df = load_data(args.input)
df = basic_clean(df)
df = feature_engineer(df)
features = prepare_features(df, save_encoder_path=args.save_encoder)
features.to_parquet(args.out_features)
print('Saved features to', args.out_features)