Deep Generative Symbolic Regression with Monte-Carlo-Tree-Search

Authors: Pierre-Alexandre Kamienny, Guillaume Lample, Sylvain Lamprier, Marco Virgolin

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we present the results of DGSR-MCTS. We begin by studying the performance on test synthetic datasets. Then, we present results on the SRBench datasets.
Researcher Affiliation Collaboration 1Meta AI, Paris, France 2ISIR MLIA, Sorbonne Universit e, France 3LERIA, Universit e d Angers, France 4Centrum Wiskunde & Informatica, the Netherlands.
Pseudocode No The paper describes its MCTS process in detail but does not provide structured pseudocode or an algorithm block.
Open Source Code No The paper does not provide an explicit statement or link indicating that the source code for its methodology is openly available.
Open Datasets Yes We evaluate DGSR-MCTS on the regression datasets of the SRBench benchmark (La Cava et al., 2021)
Dataset Splits Yes Each dataset is split into 75% training data and 25% test data using sampling with a random seed (we use 3 seeds per dataset, giving a total of 528 datasets).
Hardware Specification No The paper mentions "using 4 trainers (1 GPU/CPU each), 4 MCTS workers (1 GPU/CPU each)" but does not specify the exact models of GPUs or CPUs used.
Software Dependencies No The paper mentions the use of 'Sym Py' and 'Broyden Fletcher Goldfarb Shanno algorithm (BFGS)' but does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes The specific decoding parameters and the distribution utilized are as follows: Number of samples K per expansion. Distribution: uniform on range [8,16]. Temperature used for decoding. Distribution: uniform on range [0.5, 1.0]. Length penalty: length penalty used for decoding. Distribution: uniform on range [0, 1.2]. Depth penalty: an exponential value decay during the backup-phase, decaying with depth to favor breadth or depth. Distribution: uniform on discrete values [0.8, 0.9, 0.95, 1]. Exploration: the exploration constant puct. 1