Sort multiple `by` columns with equation in key

Would like to sort a df with an equation in the key as follows:

import pandas as pd
import numpy as np
df = pd.DataFrame({'name': ['x', 'y', 'z'], 
                   'price': [1, 2, np.nan], 
                   'cost': [3, np.nan, 1]})

df.sort_values(by=['price', 'cost'], key=lambda x: x.price/x.cost)
…doesn’t work and gives the error AttributeError: 'Series' object has no attribute 'price'

What’s the right way?

>Solution :

Better compute the ratio, sort this and reindex:

out = df.loc[df['price'].div(df['cost']).sort_values().index]

Or, if you have tricky indices, with numpy.argsort:

out = df.iloc[np.argsort(df['price'].div(df['cost']))]

Output:

  name  price  cost
0    x    1.0   3.0
1    y    2.0   NaN
2    z    NaN   1.0