Getting slow response form google OR tools

[hrkachhia]

OR tools version : 9.6.2534
Language : Python

I am using the OR tool Python library to resolve the cutting stock problem. But it has been taking too long for the last few days to respond. It is taking time more than 1 minute to respond.

I am using Python code as an API and passing the below data.

API Data :

{
    "child_rolls": [
        {
            "roll": 3,
            "size": 1490
        },
        {
            "roll": 8,
            "size": 1490
        },
        {
            "roll": 23,
            "size": 1500
        },
        {
            "roll": 9,
            "size": 1500
        },
        {
            "roll": 8,
            "size": 1480
        },
        {
            "roll": 30,
            "size": 1480
        }
    ],
    "parent_rolls": [
        {
            "roll": null,
            "size": 4470
        }
    ]
}

When I passed the above data in API it is taking more than 1 minute to return the output of the above data. This means when I pass large data to my GCP Python script gives me a slow response. So, I need your help to improve performance for all types of data. Could you please guide me and help to resolve this issue?

Python code:

import pdb
import math
from ortools.linear_solver import pywraplp
from math import ceil
from random import randint
import json
# from read_lengths import get_data
# import typer
from typing import Optional
from flask import Flask
from flask_cors import CORS
from flask import request



def newSolver(name, integer=False):
    return pywraplp.Solver(name,
                           pywraplp.Solver.CBC_MIXED_INTEGER_PROGRAMMING
                           if integer else
                           pywraplp.Solver.GLOP_LINEAR_PROGRAMMING)


'''
return a printable value
'''


def SolVal(x):
    if type(x) is not list:
        return 0 if x is None \
            else x if isinstance(x, (int, float)) \
            else x.SolutionValue() if x.Integer() is False \
                    else int(x.SolutionValue())
    elif type(x) is list:
        return [SolVal(e) for e in x]


def ObjVal(x):
    return x.Objective().Value()


def gen_data(num_orders):
    R = []  # small rolls
    # S=0 # seed?
    for i in range(num_orders):
        R.append([randint(1, 12), randint(5, 40)])
    return R


def solve_model(demands, parent_width=100,parent_main_width=0):
    '''
        demands = [
            [1, 3], # [quantity, width]
            [3, 5],
            ...
        ]

        parent_width = integer
    '''
    num_orders = len(demands)
    solver = newSolver('Cutting Stock', True)
    k, b = bounds(demands, parent_width)

    # array of boolean declared as int, if y[i] is 1,
    # then y[i] Big roll is used, else it was not used
    y = [solver.IntVar(0, 1, f'y_{i}') for i in range(k[1])]

    # x[i][j] = 3 means that small-roll width specified by i-th order
    # must be cut from j-th order, 3 tmies
    x = [[solver.IntVar(0, b[i], f'x_{i}_{j}') for j in range(k[1])]
         for i in range(num_orders)]

    # unused_widths = [solver.NumVar(0, parent_width, f'w_{j}')
    #                  for j in range(k[1])]

    unused_widths = []
    for j in range(k[1]):
        print("J",j)
        var_name = f'w_{j}'
        var = solver.NumVar(0, parent_width, var_name)

        unused_widths.append(var)
      
        print(f"Created variable {var_name} with domain [0, {parent_width}] and value {var}.")

    #print("SolVal",SolVal(unused_widths))
  
    # will contain the number of big rolls used
    nb = solver.IntVar(k[0], k[1], 'nb')
    print('1st of nb when intilize',nb)

    # consntraint: demand fullfilment
    for i in range(num_orders):
        # small rolls from i-th order must be at least as many in quantity
        # as specified by the i-th order
        solver.Add(sum(x[i][j] for j in range(k[1])) >= demands[i][0])

    # constraint: max size limit
    for j in range(k[1]):
        # total width of small rolls cut from j-th big roll,
        # must not exceed big rolls width
        solver.Add(
            sum(demands[i][1]*x[i][j] for i in range(num_orders))
            <= parent_width*y[j]
        )

        # width of j-th big roll - total width of all orders cut from j-th roll
        # must be equal to unused_widths[j]
        # So, we are saying that assign unused_widths[j] the remaining width of j'th big roll
        solver.Add(parent_main_width*y[j] - sum(demands[i][1]*x[i][j]
                   for i in range(num_orders)) == unused_widths[j])

        '''
    Book Author's note from page 201:
    [the following constraint]  breaks the symmetry of multiple solutions that are equivalent 
    for our purposes: any permutation of the rolls. These permutations, and there are K! of 
    them, cause most solvers to spend an exorbitant time solving. With this constraint, we 
    tell the solver to prefer those permutations with more cuts in roll j than in roll j + 1. 
    The reader is encouraged to solve a medium-sized problem with and without this 
    symmetry-breaking constraint. I have seen problems take 48 hours to solve without the 
    constraint and 48 minutes with. Of course, for problems that are solved in seconds, the 
    constraint will not help; it may even hinder. But who cares if a cutting stock instance 
    solves in two or in three seconds? We care much more about the difference between two 
    minutes and three hours, which is what this constraint is meant to address
    '''
        if j < k[1]-1:  # k1 = total big rolls
            # total small rolls of i-th order cut from j-th big roll must be >=
            # totall small rolls of i-th order cut from j+1-th big roll
            solver.Add(sum(x[i][j] for i in range(num_orders))
                       >= sum(x[i][j+1] for i in range(num_orders)))

    # find & assign to nb, the number of big rolls used
    solver.Add(nb == solver.Sum(y[j] for j in range(k[1])))
    print('2nd nb while executing',nb)

    ''' 
    minimize total big rolls used
    let's say we have y = [1, 0, 1]
    here, total big rolls used are 2. 0-th and 2nd. 1st one is not used. So we want our model to use the 
    earlier rolls first. i.e. y = [1, 1, 0]. 
    The trick to do this is to define the cost of using each next roll to be higher. So the model would be
    forced to used the initial rolls, when available, instead of the next rolls.

    So instead of Minimize ( Sum of y ) or Minimize( Sum([1,1,0]) )
    we Minimize( Sum([1*1, 1*2, 1*3]) )
  '''

    '''
  Book Author's note from page 201:

  There are alternative objective functions. For example, we could have minimized the sum of the waste. This makes sense, especially if the demand constraint is formulated as an inequality. Then minimizing the sum of waste Chapter 7  advanCed teChniques
  will spend more CPU cycles trying to find more efficient patterns that over-satisfy demand. This is especially good if the demand widths recur regularly and storing cut rolls in inventory to satisfy future demand is possible. Note that the running time will grow quickly with such an objective function
  '''

    Cost = solver.Sum((j+1)*y[j] for j in range(k[1]))
    # print("Cost == ",Cost)

    solver.Minimize(Cost)
    # solver.Add(Cost)
    #print("COST DATA = ",solver.Minimize(Cost))
    status = solver.Solve()
    numRollsUsed = SolVal(nb)
   #nb, x, w, demands
    return status, \
        numRollsUsed, \
        rolls(numRollsUsed, SolVal(x), SolVal(unused_widths), demands), \
        SolVal(unused_widths), \
        solver.WallTime()


def bounds(demands, parent_width=100):
    '''
    b = [sum of widths of individual small rolls of each order]
    T = local var. stores sum of widths of adjecent small-rolls. When the width reaches 100%, T is set to 0 again.
    k = [k0, k1], k0 = minimum big-rolls requierd, k1: number of big rolls that can be consumed / cut from
    TT = local var. stores sum of widths of of all small-rolls. At the end, will be used to estimate lower bound of big-rolls
    '''
    num_orders = len(demands)
    b = []
    T = 0
    k = [0, 1]
    TT = 0

    for i in range(num_orders):
        # q = quantity, w = width; of i-th order
        quantity, width = demands[i][0], demands[i][1]
        # TODO Verify: why min of quantity, parent_width/width?
        # assumes widths to be entered as percentage
        # int(round(parent_width/demands[i][1])) will always be >= 1, because widths of small rolls can't exceed parent_width (which is width of big roll)
        b.append( min(demands[i][0], int(round(parent_width / demands[i][1]))) )
        #b.append(min(quantity, int(round(parent_width / width))))

        # if total width of this i-th order + previous order's leftover (T) is less than parent_width
        # it's fine. Cut it.
        if T + quantity*width <= parent_width:
            T, TT = T + quantity*width, TT + quantity*width
        # else, the width exceeds, so we have to cut only as much as we can cut from parent_width width of the big roll
        else:
            while quantity:
                if T + width <= parent_width:
                    T, TT, quantity = T + width, TT + width, quantity-1
                else:
                    k[1], T = k[1]+1, 0  # use next roll (k[1] += 1)
    k[0] = int(round(TT/parent_width+0.5))

    print('k', k)
    print('b', b)

    return k, b


'''
  nb: array of number of rolls to cut, of each order
  
  w: 
  demands: [
    [quantity, width],
    [quantity, width],
    [quantity, width],
  ]
'''


def rolls(nb, x, w, demands):
    consumed_big_rolls = []
    num_orders = len(x)
    # go over first row (1st order)
    # this row contains the list of all the big rolls available, and if this 1st (0-th) order
    # is cut from any big roll, that big roll's index would contain a number > 0
    for j in range(len(x[0])):
        # w[j]: width of j-th big roll
        # int(x[i][j]) * [demands[i][1]] width of all i-th order's small rolls that are to be cut from j-th big roll
        h=0
        print ("int(x[i][j]",int(x[h][j])) 
        h=h+1
        RR = [abs(w[j])] + [int(x[i][j])*[demands[i][1]] for i in range(num_orders)
                            if x[i][j] > 0]  # if i-th order has some cuts from j-th order, x[i][j] would be > 0
        print("RR ",RR)
        consumed_big_rolls.append(RR)

    return consumed_big_rolls


'''
this model starts with some patterns and then optimizes those patterns
'''


def solve_large_model(demands, parent_width=100):
    num_orders = len(demands)
    iter = 0
    patterns = get_initial_patterns(demands)
    # print('method#solve_large_model, patterns', patterns)

    # list quantities of orders
    quantities = [demands[i][0] for i in range(num_orders)]
    print('quantities', quantities)

    while iter < 20:
        status, y, l = solve_master(
            patterns, quantities, parent_width=parent_width)
        iter += 1

        # list widths of orders
        widths = [demands[i][1] for i in range(num_orders)]
        new_pattern, objectiveValue = get_new_pattern(
            l, widths, parent_width=parent_width)

        # print('method#solve_large_model, new_pattern', new_pattern)
        # print('method#solve_large_model, objectiveValue', objectiveValue)

        for i in range(num_orders):
            # add i-th cut of new pattern to i-thp pattern
            patterns[i].append(new_pattern[i])

    status, y, l = solve_master(
        patterns, quantities, parent_width=parent_width, integer=True)

    return status, \
        patterns, \
        y, \
        rolls_patterns(patterns, y, demands, parent_width=parent_width)


'''
Dantzig-Wolfe decomposition splits the problem into a Master Problem MP and a sub-problem SP.

The Master Problem: provided a set of patterns, find the best combination satisfying the demand

C: patterns
b: demand
'''


def solve_master(patterns, quantities, parent_width=100, integer=False):
    title = 'Cutting stock master problem'
    num_patterns = len(patterns)
    n = len(patterns[0])
    # print('**num_patterns x n: ', num_patterns, 'x', n)
    # print('**patterns recived:')
    # for p in patterns:
    #   print(p)

    constraints = []

    solver = newSolver(title, integer)

    # y is not boolean, it's an integer now (as compared to y in approach used by solve_model)
    y = [solver.IntVar(0, 1000, '') for j in range(n)]  # right bound?
    # minimize total big rolls (y) used
    Cost = sum(y[j] for j in range(n))
    solver.Minimize(Cost)

    # for every pattern
    for i in range(num_patterns):
        # add constraint that this pattern (demand) must be met
        # there are m such constraints, for each pattern
        constraints.append(solver.Add(
            sum(patterns[i][j]*y[j] for j in range(n)) >= quantities[i]))

    status = solver.Solve()
    y = [int(ceil(e.SolutionValue())) for e in y]

    l = [0 if integer else constraints[i].DualValue()
         for i in range(num_patterns)]
    # sl =  [0 if integer else constraints[i].name() for i in range(num_patterns)]
    # print('sl: ', sl)

    # l =  [0 if integer else u[i].Ub() for i in range(m)]
    toreturn = status, y, l
    # l_to_print = [round(dd, 2) for dd in toreturn[2]]
    # print('l: ', len(l_to_print), '->', l_to_print)
    # print('l: ', toreturn[2])
    return toreturn


def get_new_pattern(l, w, parent_width=100):
    solver = newSolver('Cutting stock sub-problem', True)
    n = len(l)
    new_pattern = [solver.IntVar(0, parent_width, '') for i in range(n)]

    # maximizes the sum of the values times the number of occurrence of that roll in a pattern
    Cost = sum(l[i] * new_pattern[i] for i in range(n))
    solver.Maximize(Cost)

    # ensuring that the pattern stays within the total width of the large roll
    solver.Add(sum(w[i] * new_pattern[i] for i in range(n)) <= parent_width)

    status = solver.Solve()
    return SolVal(new_pattern), ObjVal(solver)


'''
the initial patterns must be such that they will allow a feasible solution, 
one that satisfies all demands. 
Considering the already complex model, let’s keep it simple. 
Our initial patterns have exactly one roll per pattern, as obviously feasible as inefficient.
'''


def get_initial_patterns(demands):
    num_orders = len(demands)
    return [[0 if j != i else 1 for j in range(num_orders)]
            for i in range(num_orders)]


def rolls_patterns(patterns, y, demands, parent_width=100):
    R, m, n = [], len(patterns), len(y)

    for j in range(n):
        for _ in range(y[j]):
            RR = []
            for i in range(m):
                if patterns[i][j] > 0:
                    RR.extend([demands[i][1]] * int(patterns[i][j]))
            used_width = sum(RR)
            R.append([parent_width - used_width, RR])

    return R


'''
checks if all small roll widths (demands) smaller than parent roll's width
'''


def checkWidths(demands, parent_width):
    for quantity, width in demands:
        if width > parent_width:
            print(
                f'Small roll width {width} is greater than parent rolls width {parent_width}. Exiting')
            return False
    return True


'''
    params
        child_rolls: 
            list of lists, each containing quantity & width of rod / roll to be cut
            e.g.: [ [quantity, width], [quantity, width], ...]
        parent_rolls: 
            list of lists, each containing quantity & width of rod / roll to cut from
            e.g.: [ [quantity, width], [quantity, width], ...]
'''


def StockCutter1D(child_rolls, parent_rolls, output_json=True, large_model=True):

    # at the moment, only parent one width of parent rolls is supported
    # quantity of parent rolls is calculated by algorithm, so user supplied quantity doesn't matter?
    # TODO: or we can check and tell the user the user when parent roll quantity is insufficient
    # pdb.set_trace()
    parent_width_main = parent_rolls[0][1]
    qua_arr = []
    for item in range(len(child_rolls)):
        qua_arr.append(child_rolls[item][1])
    #     # print(item)
    # # print(qua_arr)
    # # 
    if min(qua_arr) <=20:
        parent_width_second = min(qua_arr)*7
    else: 
        parent_width_second = parent_width_main


    if parent_width_second > parent_width_main:
        parent_width = parent_width_main
    else:
        parent_width = parent_width_second
    
    print("parent_width_main==",parent_width_main)
    print("parent_width_second==",parent_width_second)
    print("parent_width==",parent_width)
    if not checkWidths(demands=child_rolls, parent_width=parent_width):
        return []

    print('child_rolls', child_rolls)
    print('parent_rolls', parent_rolls)

    if not large_model:
        print('Running Small Model...')
        status, numRollsUsed, consumed_big_rolls, unused_roll_widths, wall_time = \
            solve_model(demands=child_rolls, parent_width=parent_width,parent_main_width=parent_width_main)

        # convert the format of output of solve_model to be exactly same as solve_large_model
        print('consumed_big_rolls before adjustment: ', consumed_big_rolls)
        new_consumed_big_rolls = []
        for big_roll in consumed_big_rolls:
            if len(big_roll) < 2:
                # sometimes the solve_model return a solution that contanis an extra [0.0] entry for big roll
                consumed_big_rolls.remove(big_roll)
                continue
            unused_width = big_roll[0]
            subrolls = []
            for subitem in big_roll[1:]:
                if isinstance(subitem, list):
                    # if it's a list, concatenate with the other lists, to make a single list for this big_roll
                    subrolls = subrolls + subitem
                else:
                    # if it's an integer, add it to the list
                    subrolls.append(subitem)
            new_consumed_big_rolls.append([unused_width, subrolls])
        print('consumed_big_rolls after adjustment: ', new_consumed_big_rolls)
        consumed_big_rolls = new_consumed_big_rolls

    else:
        print('Running Large Model...')
        status, A, y, consumed_big_rolls = solve_large_model(
            demands=child_rolls, parent_width=parent_width)

    numRollsUsed = len(consumed_big_rolls)
    # print('A:', A, '\n')
    # print('y:', y, '\n')

    STATUS_NAME = ['OPTIMAL',
                   'FEASIBLE',
                   'INFEASIBLE',
                   'UNBOUNDED',
                   'ABNORMAL',
                   'NOT_SOLVED'
                   ]

    output = {
        "statusName": STATUS_NAME[status],
        "numSolutions": '1',
        "numUniqueSolutions": '1',
        "numRollsUsed": numRollsUsed,
        "solutions": consumed_big_rolls  # unique solutions
    }

    # print('Wall Time:', wall_time)
    print('numRollsUsed', numRollsUsed)
    print('Status:', output['statusName'])
    print('Solutions found :', output['numSolutions'])
    print('Unique solutions: ', output['numUniqueSolutions'])

    if output_json:
        return output
    else:
        return consumed_big_rolls


'''
Draws the big rolls on the graph. Each horizontal colored line represents one big roll.
In each big roll (multi-colored horizontal line), each color represents small roll to be cut from it.
If the big roll ends with a black color, that part of the big roll is unused width.

TODO: Assign each child roll a unique color
'''

if __name__ == '__main__':

    app = Flask(__name__)
    # Enable the CORS
    CORS(app, resources={r"/*": {"origins": "*"}})

    @app.route("/")
    def index():
        return "This is index page, please go ahead...! "

    @app.route("/track", methods=['POST'])
    def track():
        try:
            print('request.data==', request.get_json())
            req_data = request.get_json()
            child_rolls = []
            parent_rolls = []
            child_rolls_obj = req_data.get('child_rolls', [])
            for item in child_rolls_obj:
                data = list(item.values())
                # data.reverse()
                child_rolls.append(data)
            parent_rolls_obj = req_data.get('parent_rolls', [])
            parent_rolls.append(list(parent_rolls_obj[0].values()))
            # pdb.set_trace()
            # number_of_cut = req_data.get('number_of_cut', [])
            if(not(len(child_rolls) and len(parent_rolls))):
                return {"success": False, "message": "Request data is not enough!"}
            consumed_big_rolls = StockCutter1D(
                child_rolls, parent_rolls, output_json=True, large_model=False)
            # return {"success": True, "data": {"solutions": consumed_big_rolls}}
            return consumed_big_rolls
        except Exception as error:
            return {"success": False, "message": "Something went wrong!", "error": error}
            # raise error

    app.config["FLASK_APP"] = 'main.py'
    app.config["FLASK_ENV"] = 'development'

    app.run(host="0.0.0.0", debug=True, port=5000, use_reloader=False)