llm-srbench/phys_osc/PO40/metadata.yaml

dataset:
  name: llm-srbench__phys_osc__PO40
  description: null
  splits:
    train:
      file: train.csv
      samples: 3500
      ratio: 7
    valid:
      file: valid.csv
      samples: 500
      ratio: 1
    id_test:
      file: id_test.csv
      samples: 500
      ratio: 1
    ood_test:
      file: ood_test.csv
      samples: 500
      ratio: 1
  ground_truth_formula:
    file: formula.py
  features:
    - name: x
      type: continuous
      train_range: [-1.2411847114562988, 1.080796480178833]
      ood_range: [-1.257206916809082, 1.0867908000946045]
      description: Position at time t
    - name: t
      type: continuous
      train_range: [0.0, 35.99919891357422]
      ood_range: [36.0072021484375, 40.0]
      description: Time
    - name: v
      type: continuous
      train_range: [-3.474342107772827, 3.478935956954956]
      ood_range: [-3.464468717575073, 3.4794163703918457]
      description: Velocity at time t
  target:
    name: dv_dt
    type: continuous
    train_range: [-11.779383659362791, 14.41136646270752]
    ood_range: [-11.81948184967041, 14.649964332580566]
    description: Acceleration in Nonl-linear Harmonic Oscillator
  license: MIT
  citation:
    arXiv: 2504.10415
    title: 'LLM-SRBench: A New Benchmark for Scientific Equation Discovery with Large Language Models'
  resources:
    - https://github.com/deep-symbolic-mathematics/llm-srbench
    - https://huggingface.co/datasets/nnheui/llm-srbench