Large-Scale Price Optimization via Network Flow

Authors: Shinji Ito, Ryohei Fujimaki

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical results show that the proposed algorithm can successfully handle hundreds of products and derive solutions as good as other state-of-the-art methods, while its computational cost is much cheaper, regardless of whether the substitute-goods property holds or not.
Researcher Affiliation Industry Shinji Ito NEC Corporation s-ito@me.jp.nec.com Ryohei Fujimaki NEC Corporation rfujimaki@nec-labs.com
Pseudocode Yes Algorithm 1 s-t cut for price optimization with the substitute-goods property
Open Source Code No No explicit statement or link for open-source code for the proposed method is provided.
Open Datasets No We applied the proposed method to actual retail data from a middle-size supermarket located in Tokyo [23].10 The Data has been provided by KSP-SP Co., LTD, http://www.ksp-sp.com. While the provider is mentioned, a specific link, DOI, or formal citation for public access to the dataset is not provided.
Dataset Splits No The paper states: 'we used the first 35 months (1065 samples) for training regression models and simulated the best price strategy for the next 20 days.' This describes a training period and a future simulation period but does not explicitly define a separate validation split or discuss cross-validation.
Hardware Specification Yes All experiments were conducted in a machine equipped with Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz, 768GB RAM.
Software Dependencies Yes We use SDPA 7.3.8 to solve SDP problems7 and use the implementation of QPBO and QPBOI written by Kolmogolov.8
Experiment Setup Yes The sales quantity qi of the i-th product was generated from the regression model qi = αi + PM j=1 βijpj, where {αi} and {βij} were generated by uniform distributions. We considered three types of uniform distributions to investigate the effect of submodularity, as shown in Table 1, which correspond to three different situations: (i) all pairs of products are substitute goods, i.e., the gross profit function is supermodular, (ii) half pairs are substitute goods and the others are complementary goods, i.e., the gross profit function contains submodular terms and supermodular terms, and (iii) all pairs are complementary goods, i.e., the gross profit function is submodular. Price candidates Pi and cost ci for each product are fixed to Pi = {0.6, 0.7, . . . , 1.0} and ci = 0, respectively.