TripleTree: A Versatile Interpretable Representation of Black Box Agents and their Environments

Authors: Tom Bewley, Jonathan Lawry11415-11422

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We initially validate TRIPLETREE in a simple MDP with 2 state features and 2 discrete actions... Figure 3 shows the result of growing a TRIPLETREE of up to 200 leaves using these four datasets, with various impurity weightings θ... Figure 7 shows how the three losses vary during growth on both the training set and a validation set.
Researcher Affiliation Academia Tom Bewley and Jonathan Lawry Department of Engineering Mathematics, University of Bristol, Bristol, United Kingdom {tom.bewley, j.lawry}@bristol.ac.uk
Pseudocode No The paper describes the algorithms and their modifications using text and mathematical equations, but it does not provide formal pseudocode or algorithm blocks.
Open Source Code Yes A Python implementation of TRIPLETREE is available on Git Hub at https://github.com/tombewley/Triple Tree.
Open Datasets No For each, we create a dataset D with 10^4 samples, by running randomly-initialised episodes of 100 timesteps. Using a dataset of 10^5 observations, we grow a TRIPLETREE....The paper describes generating its own datasets and does not provide specific access information (link, DOI, or citation with authors/year for a public dataset) for these datasets. While OpenAI Gym is mentioned, the generated observations are not made available.
Dataset Splits No Figure 7 shows how the three losses vary during growth on both the training set and a validation set. The paper mentions training and validation sets but does not provide specific details on the dataset splits (e.g., percentages or sample counts) needed for reproduction.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions 'A Python implementation' and 'Open AI Gym' and cites 'Baselines Zoo', but does not provide specific version numbers for any of these software dependencies or other libraries.
Experiment Setup Yes For a given Rleft, Rright, Rspeed, discount factor γ (we use γ = 0.99), and suitable discretisation of S (we use a 30 30 grid)...For each, we create a dataset D with 10^4 samples, by running randomly-initialised episodes of 100 timesteps.Using a dataset of 10^5 observations, we grow a TRIPLETREE of up to 1000 leaves with θ = [1, 1, 1]...We use the validation losses to inform early stopping and select the 450-leaf tree for evaluation.