Counterfactual Strategies for Markov Decision Processes

Authors: Paul Kobialka, Lina Gerlach, Francesco Leofante, Erika Ábrahám, Silvia Lizeth Tapia Tarifa, Einar Broch Johnsen

IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our approach on four real-world datasets and demonstrate its practical viability in sophisticated sequential decision-making tasks.
Researcher Affiliation Academia 1University of Oslo, Oslo, Norway 2RWTH Aachen University, Germany 3Imperial College London, United Kingdom EMAIL, EMAIL, EMAIL
Pseudocode No The paper describes the optimization problem using mathematical formulations (Constraints (1)-(13)) and prose, but does not include any clearly labeled pseudocode or algorithm blocks. The steps are described in regular paragraph text without structured formatting that would constitute pseudocode.
Open Source Code No The paper does not provide an explicit statement about releasing source code, a direct link to a code repository, or mention of code in supplementary materials for the methodology described. It references an extended version on arXiv, but this is not a code release statement.
Open Datasets Yes In our experiments, we consider four real-world datasets. Grep S records customer interaction with a programming skill evaluation service [Kobialka et al., 2022]. BPIC12 [van Dongen, 2012] and BPIC17 [van Dongen, 2017], which record the loan application procedure in a bank, stem from the Business Process Intelligence Challenge2 of the IEEE Task Force on Process Mining.3 MSSD is the Music Streaming Sessions Dataset [Brost et al., 2019] from Spotify; we consider the small version of MSSD, with 10 000 listening sessions.
Dataset Splits No The paper mentions subsets of the MSSD dataset (e.g., MSSD10, MSSD40 representing 10% and 40% of the dataset) but does not provide specific training/test/validation dataset splits, exact percentages, sample counts, or explicit splitting methodology for reproducing experiments. It only describes how models were constructed based on data volume.
Hardware Specification No The paper describes the experimental setup and evaluation of the method's performance but does not provide specific details about the hardware used to run the experiments, such as GPU/CPU models, memory, or cloud instance types.
Software Dependencies No The paper does not provide specific software dependencies with version numbers, such as programming languages, libraries, or solvers with their exact versions, that would be needed to replicate the experiments.
Experiment Setup Yes We randomly generate ten initial user strategies for each model and let the target probability ̓ range over {0.0001} {0.1, 0.2, . . . , 1}, where 0.0001 represents near-perfect performance. In this work, we use r0 = r1 = r = 1 and ̓ = 2 to weight each distance component equally and to weight diversity higher than distances.