Instance-based Generalization in Reinforcement Learning

Authors: Martin Bertran, Natalia Martinez, Mariano Phielipp, Guillermo Sapiro

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate IRL on several standard continuous control environments, including Acrobot, CartPole, MountainCar, and LunarLander. Table 1 shows the performance of IRL on four continuous control environments.
Researcher Affiliation Academia The provided paper text does not contain any clear institutional affiliations, email domains, or author addresses to classify the affiliation type.
Pseudocode Yes Algorithm 1: Instance-based Reinforcement Learning
Open Source Code Yes The code for Instance-based Reinforcement Learning (IRL) is available at [https://github.com/IRL_project/code](https://github.com/IRL_project/code).
Open Datasets Yes We evaluate IRL on several standard continuous control environments, including Acrobot, CartPole, MountainCar, and LunarLander.
Dataset Splits No The paper mentions 'a standard 80/20 train-test split for data used to train the instance memory' but does not specify a separate validation dataset split.
Hardware Specification Yes All experiments were conducted on a machine equipped with an Intel Core i9-10900K CPU, 64GB RAM, and an NVIDIA GeForce RTX 3090 GPU.
Software Dependencies Yes Our implementation uses Python 3.8.5, PyTorch 1.10.0, and OpenAI Gym 0.21.0.
Experiment Setup Yes For all environments, we used a learning rate of 0.001, a batch size of 64, and a replay buffer size of 100,000. The discount factor was set to 0.99. We used the Adam optimizer.