# Sum Operations

operation_fields = [column for column in data.columns if 'Operations' in column]

data['Operations'] = data[operation_fields].apply(lambda row: sum(row), axis=1)

for operation_field in operation_fields:

    data[f'Operations_{operation_field}_vs_total'] = np.where(

        data['Operations'] == 0,

        np.floor(data['Operations']),

        np.floor(data[operation_field] / data['Operations'] * 100)

    )

# Sum transactions

transaction_fields = [column for column in data.columns if 'Transactions' in column and 'SavingAccount' in column]

data['Transactions'] = data[transaction_fields].apply(lambda row: sum(row), axis=1)

for transaction_field in transaction_fields:

    data[f'Transactions_{transaction_fields}_vs_total'] = np.where(

        data['Transactions'] == 0,

        np.floor(data['Transactions']),

        np.floor(data[transaction_fields] / data['Transactions'] * 100)

    )

# vectorized sum, which is more memory efficient

data['Transactions'] = data[transaction_fields].sum(axis=1)

# numpy to avoid the explicit loop

transaction_ratios = data[transaction_fields].div(data['Transactions'], axis=0)

transaction_ratios = transaction_ratios.multiply(100).fillna(0).astype(int)

for transaction_field in transaction_fields:

    data[f'Transactions_{transaction_field}_vs_total'] = transaction_ratios[transaction_field]