sql cube to pandas

from sqlalchemy import create_engine
import pandas as pd
import numpy as np
import functools

orig_df = None
def get_cube(*fields):
    global orig_df
    # fields that we use in SQL group by rollup
    all_fields = ['transit_time', 'rounded_area', 'epoch', 'year', 'city_id']
    if set(fields) - set(all_fields):
        raise ValueError('fields %s are not in the columns' % ', '.join(set(fields) - set(all_fields)))

    # we store the original dataframe to get new groups again
    # this is purely procedural and stored in global space, as I have the same cube in the entire project
    if orig_df is None:
        engine = create_engine('postgresql://localhost:5432/alcohol')
        orig_df = pd.read_sql_query('select * from "cube_stat"', engine)

    df = orig_df
    # in cube, fields that we omit have to be null
    for f in (set(all_fields) - set(fields)):
        df = df.loc[pd.isnull(df[f])]

    # fields that we iterate on must be not null
    for f in fields:
        df = df[df[f].notnull()]

    return df

# same but a standard iterator over cities to build charts
def iter_cities(*fields):
    fields = set(fields) | set(['city_id'])
    cube = get_cube(*list(fields))

    for city_name, group in cube.groupby('name'):
        if city_name is None:
            continue

        df1 = pd.DataFrame(group)
        if len(df1) == 0:
            continue

        yield city_name, df1