An approximate dynamic programming approach for multi-stage stochastic lot-sizing under a Decision-Hazard-Decision information structure

This dataset contains anonymized data for several instances of the capacitated lot-sizing problem with inventory bounds and lost sales. The dataset was created in January 2025 based on the research activities of the publisher's article. The dataset is stored in two separate .zip files. A zip file contains multiple .json files that each corresponds to the data for twenty-five instances sharing similar metrics. The dataset data_simple represents instances with a large number of stages and a small number of time periods and scenarios per stage. The dataset data_complex represents instances with a small number of stages and a large number of time periods and scenarios per stage. The dataset data_simple and data_complex present respectively sixteen and two classes of instances exploring various configurations of costs and capacities. This data was generated to model multi-stage stochastic lot-sizing problems but it can be used in a deterministic setting by ignoring all information related to the presence of uncertainty. All informations related to the generation and use of data to represent problem instances can be found in the articleV. Spitzer, C. Gicquel, E. Gurevsky and F. Sanson. An approximate dynamic programming approach for multi-stage stochastic lot-sizing under a Decision-Hazard-Decision information structure. Discrete Applied Mathematics, 2026, 379, pp.355-378.