Skip to content

Instantly share code, notes, and snippets.

@xpl
Last active December 18, 2019 09:16
Show Gist options
  • Save xpl/f93af7ea2d2bd08fad2c45535433327d to your computer and use it in GitHub Desktop.
Save xpl/f93af7ea2d2bd08fad2c45535433327d to your computer and use it in GitHub Desktop.
Mixed Integer Programming Example In Python

Mixed Integer Programming Example (Python)

import pandas as pd
import numpy  as np

from scipy.spatial import distance
from mip.model     import *

factories = pd.DataFrame ([
    ['F_01', 59.250276, 59.871183, 389],
    ['F_02', 84.739320, 14.179336, 409],
    ['F_03', 42.397937, 42.474530, 124],
    ['F_04', 19.539202, 13.714643, 70],
    ['F_05', 41.280669, 37.860993, 386],
    ['F_06', 37.159066, 41.353602, 196],
    ['F_07', 96.890453, 64.420010, 394],
    ['F_08', 86.267499, 81.662811, 365],
    
], columns=['fact', 'x', 'y', 'supply']).set_index ('fact')

shops = pd.DataFrame ([
    ['S_01', 13.490869, 73.269974, 200],
    ['S_02', 85.435435, 66.637250, 20],
    ['S_03', 28.578297, 8.997380, 320],
    ['S_04', 31.324145, 91.839907, 360],
    ['S_05', 40.338575, 15.487028, 360],
    ['S_06', 41.642451, 42.121572, 120],
    ['S_07', 53.983692, 20.950457, 360],
    ['S_08', 75.761895, 87.067552, 60],
    ['S_09', 81.836739, 36.799647, 80],
    ['S_10', 54.260517, 25.920108, 100],
    ['S_11', 67.918105, 68.108601, 340],
    ['S_12', 92.200710, 10.898110, 360],
    ['S_13', 19.966539, 39.046271, 60],
    
], columns=['shop', 'x', 'y', 'demand']).set_index ('shop')

unit_revenue              = 1200
unit_production_cost      = 200
unit_delivery_cost_per_km = 20
shop_rent_cost            = 100000

N_facts = len (factories)
N_shops = len (shops)

to_df = lambda arr: pd.DataFrame (arr, index=shops.index, columns=factories.index)

factories_vs_shops = lambda fn: [[*(fn (f, s) for _, f in factories.iterrows ())] for _, s in shops.iterrows ()]
unit_delivery_cost = lambda f, s: (distance.euclidean ((f.x, f.y), (s.x, s.y)) * unit_delivery_cost_per_km)
unit_profit        = lambda f, s: unit_revenue - (unit_production_cost + unit_delivery_cost (f, s))
unit_profits       = factories_vs_shops (unit_profit)

max_supply       = factories.supply.values
max_demand       = shops.demand.values
max_total_supply = np.sum (max_supply)

def display_and_check_results (units, shops_open = np.ones (N_shops)):

    supply = np.sum (units, axis=0)
    demand = np.sum (units, axis=1)
        
    df = to_df (units).replace (0.0, '')
    
    shop_profits = to_df (units * unit_profits).sum (axis=1)
    
    df['profit'] = shop_profits
    df['rent']   = np.array (shops_open) * shop_rent_cost
    df['total']  = shop_profits - df['rent']

    df.loc['']      = None
    df.loc['total'] = df.sum (axis=0)
    
    display (df.fillna (''))
    
    grand_total = df['total']['total']
    
    print (f'Grand total: {grand_total}')
    
    # All supply should be used
    assert (max_total_supply == np.sum (supply))

    # Supply & demand constraints should be satisfied!
    assert (np.alltrue (np.round (supply) == max_supply) and np.alltrue (np.round (demand) <= max_demand))
    
# --------------------- SOLVE ----------------------------

m = Model (sense=MAXIMIZE, solver_name=CBC)

profits_coeffs = np.array ([[unit_profits[s][f] for f in range (N_facts)] for s in range (N_shops)]).flatten ()

# N×M matrix of decision variables (shop-wise)
supplies_s = [[*(m.add_var (var_type=CONTINUOUS, lb=0) for _ in range (N_facts))] for _ in range (N_shops)]

# The same but factory-wise
supplies_f = np.transpose (supplies_s)

# The same but all-flat
supplies = np.array (supplies_s).flatten ()

# Binary variables encoding open/closed state for shops
shops_open = [m.add_var (var_type=BINARY) for _ in range (N_shops)]

# Add linear constraints
for s in range (N_shops): m += xsum (supplies_s[s]) <= float (max_demand[s]) * shops_open[s]
for f in range (N_facts): m += xsum (supplies_f[f]) == float (max_supply[f])

# Objective function (what we are optimizing)
m.objective = (xsum (supplies[i]   * profits_coeffs[i] for i in range (len (supplies))) -
               xsum (shops_open[i] * shop_rent_cost    for i in range (N_shops)))

# Run it!
display (m.optimize ())
display (m.objective_value)

display_and_check_results (np.array ([v.x for v in supplies]).reshape (N_shops, N_facts),
                           [v.x for v in shops_open])
<OptimizationStatus.OPTIMAL: 0>



133842.80279694987
fact F_01 F_02 F_03 F_04 F_05 F_06 F_07 F_08 profit rent total
shop
S_01 0 0 0
S_02 0 0 0
S_03 70 27 196 130476 100000 30475.9
S_04 81 279 -20565.4 100000 -120565
S_05 1 359 198671 100000 98671.3
S_06 120 117999 100000 17998.7
S_07 308 49 3 85734.3 100000 -14265.7
S_08 60 45822.8 100000 -54177.2
S_09 80 29669.9 100000 -70330.1
S_10 0 0 0
S_11 314 26 144722 100000 44722.4
S_12 360 301313 100000 201313
S_13 0 0 0
total 1.03384e+06 900000 133843
Grand total: 133842.8027969498
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment