Learning With Options That Terminate Off-Policy

Authors: Anna Harutyunyan, Peter Vrancx, Pierre-Luc Bacon, Doina Precup, Ann Nowé

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our algorithm empirically, and show that it holds up to its motivating claims.
Researcher Affiliation Collaboration Anna Harutyunyan Vrije Universiteit Brussel Brussels, Belgium Peter Vrancx PROWLER.io Cambridge, England Pierre-Luc Bacon Mc Gill University Montreal, Canada Doina Precup Mc Gill University Montreal, Canada Ann Nowe Vrije Universiteit Brussel Brussels, Belgium
Pseudocode Yes Algorithm 1 presents the forward view of the algorithm underlying this expected operator, for the general case of an evolving sequence of policies (μk)k N. This algorithm is very similar to the recently formalized Q(σ) (Asis et al. 2017; Sutton and Barto 2017), with β being a state-option generalization of 1 σ.
Open Source Code No The paper does not provide any explicit statements about releasing its own source code, nor does it provide a link to a code repository for the described methodology.
Open Datasets No The paper mentions '19-state random walk task', 'Modified Cliffwalk', and 'Pinball domain'. While these are known reinforcement learning tasks, the paper does not provide specific links, DOIs, or formal citations for accessing the exact dataset configurations or environment implementations used in their experiments.
Dataset Splits No The paper mentions setting 'step-sizes set via a linear search over α {0.1,0.2,0.3,0.4}' but does not specify any explicit dataset splits (e.g., percentages or sample counts) used for validation during this search or for training/testing.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments, such as CPU models, GPU models, or memory specifications.
Software Dependencies No The paper does not list any specific software dependencies with their version numbers (e.g., programming languages, libraries, or frameworks like PyTorch, TensorFlow, etc.) that would be necessary for reproducibility.
Experiment Setup Yes The termination conditions β and ζ are evaluated in the range of {0.1,0.5,0.8,1}, with the first value being positive to ensure adequate state visitation. The step-sizes set via a linear search over α {0.1,0.2,0.3,0.4}. The discount factor γ = 0.99.