junyaokk/ipmof

Online-Retail Inventory Placement and Multi-item Order Fulfillment (IPMOF)

This is our implementation for the paper:

• Online-Retail Inventory Placement and Multi-item Order Fulfillment

We provides the Python codes for data loading, data generation and all algorithms considered in this paper.

If you encounter any significant bugs, then please contact Junyao Yu.

Usage

We have selected two specific retailers' data, namely Retailer 1 and Retailer 2, for evaluating the performance.

To facilitate more accurate estimations of the stochastic program and test different problem sizes for our algorithms, we have developed a scenario generator for the original sets. Define the problem with generated scenarios set as a problem instance, which is served as the input for our algorithms.

The reader can specify the retailer index, replenishment frequency, and the number of scenarios to generate a problem instance.

First, we create an instance base to load and preprocess the data.

from domain.instance import InstanceBase
instance_base = InstanceBase(retailer_id='retailer1', 
                             data_dir='./data/retailer1/', 
                             scenario_unit=7)

To optimize the problem instance using the instance base, users need to specify the following parameters to create a task.

• task_id: the unique index of the task.
• need solver: whether the Gurobi solver should be invoked to compute the final cost based on the solution obtained from our algorithms (True or False).
• need opt: whether the Gurobi solver should be invoked to optimize the problem (True or False).
• need benchmark: whether the product-level benchmark should be invoked to optimize the problem (True or False).
• capacity constr: whether the problem includes a capacity constraint (True or False).
• capacity type: the type of capacity constraint, it can be 'warehouse' or 'assortment'.
• input fdc capacity: the value of the single capacity constraint.
• both paras: the values of the two types capacity constraints.
• algorithm paras: the values of parameters in our algorithms.
• scenario num: the number of scenarios within a problem instance.
• data dir: the directory for storing the task results.

The default parameters for these algorithms are specified in 'default_paras.py'. To introduce the implementation of three algorithms, they are run with the same parameters set. However, this does not guarantee the optimality of the results. For users, we provide a summary of the recommended parameter intervals and values. Further details can be found in Appendix D of our paper.

The relative test functions have been encapsulated, readers can read them in detail. For example, we create three tasks from above instance base for PBS, P&K and hybrid algorithms.

Here, we simply set the random seed to 3407. Note that under well-tuned parameters setting, the randomness of the problem instances does not affect the performance of our algorithms.

import random
from domain.task import Task
random.seed(3407)

• PH-based Bounded Subgradient (PBS) Algorithm

task_pbs = Task(task_id='TASK_001',
                instance_base=instance_base, 
                capacity_constr=True,
                capacity_type='warehouse',
                need_solver=True,
                need_opt=True,
                need_benchmark=True,
                input_fdc_capacity=8000)
task_pbs.run(scenario_num=10, data_dir='./data/retailer1/')

• PH-based Algorithm & Knapsack Problem with Dependencies (P&K)

task_pnk = Task(task_id='TASK_002',
                instance_base=instance_base, 
                capacity_constr=True,
                capacity_type='assortment',
                need_solver=True,
                need_opt=True,
                need_benchmark=True,
                input_fdc_capacity=190
                )
task_pnk.run(scenario_num=10, data_dir='./data/retailer1/')

• Hybrid Algorithm

task_hybrid = Task(task_id='TASK_003',
                instance_base=instance_base, 
                capacity_constr=True,
                capacity_type='both',
                need_solver=True,
                need_opt=True,
                need_benchmark=True,
                both_paras={'assortment': 95,
                            'warehouse': 4000}
                )
task_hybrid.run(scenario_num=10, data_dir='./data/retailer1/')