- model: cartpole-ann
  description: Train a simple ANN on the base cartpole environment
  operations:
    train:
      description: Train ANN model
      main: snnrl/algos/vpg/vpg --save_dir data
      output-scalars: no
      flags-import-skip: [save_dir]
      flags:
        device:
          default: cuda
        gym_env:
          default: "ImageCartPole-v0"
        epochs:
          default: 50
        steps_per_epoch:
          default: 4000
        max_ep_length:
          default: 1000
        policy_lr:
          default: 3e-4
        vf_lr:
          default: 1e-3
        vf_iters:
          default: 400
        gae_gamma:
          default: 0.99
        gae_lam:
          default: 0.95
        policy_hidden:
          default: "32,64"
        vf_hidden:
          default: "32,64"
      compare:
        - ep_ret_mean/train as ep_ret_mean
        - ep_ret_var/train as ep_ret_var
        - ep_len_mean/train as ep_len_mean
        - ep_len_var/train as ep_len_var
    image-diff:
      description:
        Train the same model, but using the diff environment and hope
        it is better.
      steps:
        - run: train gym_env=ImageDiffCartPole-v0

- model: cartpole-snn
  description: Train a simple SNN on the image cartpole environment
  operations:
    train:
      description: Train SNN model
      main: snnrl/algos/vpg/snn_vpg --save_dir data
      output-scalars: no
      flags-import-skip: [save_dir]
      flags:
        device:
          default: cuda
        gym_env:
          default: "ImageCartPole-v0"
        epochs:
          default: 50
        steps_per_epoch:
          default: 4000
        max_ep_length:
          default: 1000
        policy_lr:
          default: 3e-4
        vf_lr:
          default: 1e-3
        vf_iters:
          default: 400
        gae_gamma:
          default: 0.99
        gae_lam:
          default: 0.95
        snn_config:
          default: "./configs/snn.json"
      requires:
        - file: snnrl/configs
          path: .
      compare:
        - ep_ret_mean/train as ep_ret_mean
        - ep_ret_var/train as ep_ret_var
        - ep_len_mean/train as ep_len_mean
        - ep_len_var/train as ep_len_var