defaults:
  - _self_
hydra:
  run:
    dir: ${logdir}
_target_: agent.pretrain.train_gaussian_agent.TrainGaussianAgent

name: avoid_m2_pre_gmm_mlp_ta${horizon_steps}
logdir: ${oc.env:DPPO_LOG_DIR}/d3il-pretrain/${name}/${now:%Y-%m-%d}_${now:%H-%M-%S}_${seed}
train_dataset_path: ${oc.env:DPPO_DATA_DIR}/d3il/avoid_m2/train.npz

seed: 42
device: cuda:0
env: avoid
mode: d57_r12 # M2, desired modes 5 and 7, required modes 1 and 2
obs_dim: 4
action_dim: 2
horizon_steps: 4
cond_steps: 1
num_modes: 5

wandb:
  entity: ${oc.env:DPPO_WANDB_ENTITY}
  project: d3il-${env}-pretrain
  run: ${now:%H-%M-%S}_${name}

train:
  n_epochs: 10000
  batch_size: 32
  learning_rate: 1e-4
  weight_decay: 1e-6
  lr_scheduler:
    first_cycle_steps: 10000
    warmup_steps: 100
    min_lr: 1e-5
  epoch_start_ema: 20
  update_ema_freq: 10
  save_model_freq: 1000

model:
  _target_: model.common.gmm.GMMModel
  network:
    _target_: model.common.mlp_gmm.GMM_MLP
    mlp_dims: [256, 256]  # smaller MLP for less overfitting
    activation_type: ReLU
    residual_style: False
    fixed_std: 0.1
    num_modes: ${num_modes}
    cond_dim: ${eval:'${obs_dim} * ${cond_steps}'}
    horizon_steps: ${horizon_steps}
    action_dim: ${action_dim}
  horizon_steps: ${horizon_steps}
  device: ${device}

ema:
  decay: 0.995

train_dataset:
  _target_: agent.dataset.sequence.StitchedSequenceDataset
  dataset_path: ${train_dataset_path}
  horizon_steps: ${horizon_steps}
  cond_steps: ${cond_steps}
  device: ${device}