environment/config.yaml

project_name : "Quadmesh"
experiment_name : "full-dataset-obs36-10darts-v0"

description : "Apprentissage avec une observation de taille 36. Après optimisation de l'environnement. Nouvelle observatin avec dummy vertices et pas de temps augmenté à 150  "

total_timesteps : 500000

paths:
  log_dir : "training/results/e1/"
  policy_saving_dir : "training/policy_saved/e1/"
  wandb_model_saving_dir : "training/wandb_models/e1/"
  episode_recording_dir: "training/results/e1/episode_recordings/" # Not working
  observation_counts_dir: "training/results/e1/observation_counts/"
  parameters_saving_dir : "training/results/e1/parameters/"

dataset:
  training_dataset_dir : "training/dataset/training_dataset"
  exploit_dataset_dir : "training/dataset/test_dataset"
  #If datasets directory above are set to None, the agent will either train or be exploited on a single mesh file given below
  training_mesh_file_path : "mesh_files/simple_quad.msh"
  evaluation_mesh_file_path : "mesh_files/simple_quad.msh"

seed : 1

env:
  env_id : "Quadmesh-v0"
  n_vec_envs : 0
  max_episode_steps : 75
  n_darts_selected : 10
  deep : 36
  obs_size : 360 # It needs to be calculated manually
  action_restriction : false
  with_degree_observation : false
  render_mode : null
  obs_count : false
  analysis_type : "topo" #only the topo analysis type is working
  debug : false

ppo:
  policy : MlpPolicy
  n_steps : 2048
  n_epochs : 5
  batch_size: 64
  learning_rate : 0.0001
  gamma : 0.9
  clip_range : 0.2

eval:
  eval_env_id : "Quadmesh-v0"
  max_episode_steps : 300
  n_darts_selected: 5
  deep: 36
  obs_size: 180
  render_mode: null #"human"
  action_restriction: false
  with_quality_observation: false
  # The three parameters below are not used anymore
  min_evals : 5
  max_no_improvement_evals : 10
  eval_freq : 500


metrics:
  normalized_return: 0
  ep_len_mean: 0
  ep_reward_mean: 0