{
  "name": "CleanRL ppo.py command-line flags grouped by role",
  "description": "The command-line flags listed by python cleanrl/ppo.py --help in the CleanRL basic-usage documentation, grouped by the role each plays in an experiment (logging, reproducibility, rollout, optimisation, PPO objective). Grouping and one-line descriptions compiled from the documentation's flag list and standard PPO terminology.",
  "page": "https://casrai.org/guides/cleanrl",
  "publisher": "CASRAI",
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "dateModified": "2026-11-16T22:05:05Z",
  "temporalCoverage": null,
  "spatialCoverage": null,
  "columns": [
    {
      "name": "Flag",
      "unit": "",
      "description": ""
    },
    {
      "name": "Role",
      "unit": "",
      "description": ""
    },
    {
      "name": "What it controls",
      "unit": "",
      "description": ""
    }
  ],
  "rows": [
    {
      "Flag": "--exp-name",
      "Role": "Logging",
      "What it controls": "Experiment name used to label a run"
    },
    {
      "Flag": "--env-id",
      "Role": "Environment",
      "What it controls": "Environment to train on (example: CartPole-v0)"
    },
    {
      "Flag": "--seed",
      "Role": "Reproducibility",
      "What it controls": "Random seed for the run"
    },
    {
      "Flag": "--total-timesteps",
      "Role": "Budget",
      "What it controls": "Total training steps"
    },
    {
      "Flag": "--torch-deterministic",
      "Role": "Reproducibility",
      "What it controls": "PyTorch determinism option"
    },
    {
      "Flag": "--cuda",
      "Role": "Hardware",
      "What it controls": "GPU use"
    },
    {
      "Flag": "--track",
      "Role": "Logging",
      "What it controls": "Weights and Biases tracking"
    },
    {
      "Flag": "--capture_video",
      "Role": "Logging",
      "What it controls": "Saves gameplay video to videos/{run_name}"
    },
    {
      "Flag": "--learning-rate",
      "Role": "Optimisation",
      "What it controls": "Learning rate"
    },
    {
      "Flag": "--num-envs",
      "Role": "Rollout",
      "What it controls": "Number of parallel environments"
    },
    {
      "Flag": "--num-steps",
      "Role": "Rollout",
      "What it controls": "Steps collected per environment per rollout"
    },
    {
      "Flag": "--num-minibatches",
      "Role": "Optimisation",
      "What it controls": "Minibatches per update"
    },
    {
      "Flag": "--update-epochs",
      "Role": "Optimisation",
      "What it controls": "Epochs per update"
    },
    {
      "Flag": "--gamma",
      "Role": "Return",
      "What it controls": "Discount factor"
    },
    {
      "Flag": "--gae-lambda",
      "Role": "Return",
      "What it controls": "Generalised advantage estimation lambda"
    },
    {
      "Flag": "--anneal-lr",
      "Role": "Optimisation",
      "What it controls": "Learning-rate annealing option"
    },
    {
      "Flag": "--norm-adv",
      "Role": "Optimisation",
      "What it controls": "Advantage normalisation option"
    },
    {
      "Flag": "--clip-coef",
      "Role": "PPO objective",
      "What it controls": "Surrogate clipping coefficient"
    },
    {
      "Flag": "--clip-vloss",
      "Role": "PPO objective",
      "What it controls": "Value-loss clipping option"
    },
    {
      "Flag": "--ent-coef",
      "Role": "PPO objective",
      "What it controls": "Entropy bonus coefficient"
    },
    {
      "Flag": "--vf-coef",
      "Role": "PPO objective",
      "What it controls": "Value-function loss coefficient"
    },
    {
      "Flag": "--max-grad-norm",
      "Role": "Optimisation",
      "What it controls": "Gradient-norm clipping limit"
    },
    {
      "Flag": "--target-kl",
      "Role": "PPO objective",
      "What it controls": "KL-divergence target for early stopping of updates"
    }
  ],
  "sources": [
    {
      "title": "CleanRL documentation: basic usage",
      "url": "https://docs.cleanrl.dev/get-started/basic-usage/",
      "accessed": "2026-11-14"
    },
    {
      "title": "CleanRL: High-quality Single-file Implementations of Deep Reinforcement Learning Algorithms (JMLR 2022)",
      "url": "https://jmlr.org/papers/v23/21-1342.html",
      "accessed": "2026-11-14"
    }
  ]
}