Update TCFlarge_3D_Re207/environment_config.yaml

TCFlarge_3D_Re207/environment_config.yaml

@@ -1,112 +1,106 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+# PettingZoo reproducible config for TCFmini_3D_Re180
+# Intended to be used with prepare_workspace + mpirun from work_dir.
+
+runner:
+  # RL/interaction settings
+  nb_interactions: 2500
+  nb_episodes: 100
+  nb_warmup_episodes: 1
+  train_steps: 300
+  gradient_steps: 64
+  ckpt_int: 5  # Save checkpoint every 5 episodes
+  RL_algorithm: "TD3"
+  rescale_actions: true
+  learning_rate: 1e-3
+  batch_size: 2048
+  buffer_size: 20_000_000
+  learning_starts: 100
+  tau: 0.005
+  policy_delay: 2
+  target_policy_noise: 0.2
+  target_noise_clip: 0.5
+  action_noise: 0.1 # 10% of the action space
+
+  # Reproducibility and environment coupling
+  seed: 20010603 # Fixed seed for reproducibility
+  random_init: 6
+  rank: 1
+  evaluation: false
+  normalize_input: "utau"
+  ctrl_min_amp: -0.063
+  ctrl_max_amp: 0.063
+  npl_state: 2
+  u_tau: 0.0630
+  dUdy: 13.10
+
 simulation:
-  CASENAME: tcf
-  exeName: nek5000
-  nzs: 1
-  nxs: 1
-  out_log: logfile
-  host: localhost
-  hostfile: ''
-  compile_path: ../envs/cases/large_channel_noresize
-  restart_folder: ../data/rs6_lc_noresize_new_reth515
+  CASENAME: "tcf"
+  solver_version: "v19"
+  exeName: "nek5000"
+  host: "localhost"
+  hostfile: ""
+  compile_path: "."
+  restart_folder: "restart_files"
+
+  # Core runtime configuration
   nproc: 512
-  TOTCTRL: 98
-  ndrl: 25
-  znmf_avg: 1
+  TOTCTRL: 128 # Buffer size for MPI
+  ndrl: 13 # dt+ = 0.6 
+  znmf_avg: 1 
   target_cfl: 2.0
   y_sensing: 15.0
-  retau: 280.0
-  num_nodes_srun: 4
-  oversubscribe: false
-  lx1: 8
-  Lx: 6.28
+  retau: 207.0
+  # --- We are not using body-force method in this case ---
+  ys_bdf: 0.0
+  amp_bdf: 0.0
+  ret_bdf: 0.0
+
+  # SIZE (mesh)
+  lx1: 8 
+  Lx: 6.28 
   Ly: 1.0
   Lz: 3.14
   Nx: 10
-  Ny: 8
+  Ny: 9
   Nz: 9
+  # -- Size of the actuator patch --
+  nxs: 1
+  nzs: 1
+
+  # .par settings
   tmax: 100000.0
-  stopAt: numSteps
+  stopAt: "numSteps"
   numSteps: 500000
-  dt: -0.002
-  timeStepper: bdf3
-  variableDt: 'no'
-  writeControl: timeStep
-  writeInterval: 200000
-  dealiasing: 'yes'
-  filtering: none
+  dt: -4.0e-03
+  timeStepper: "bdf3"
+  variableDt: "no"
+  writeControl: "timeStep"
+  writeInterval: 100000
+  dealiasing: "yes"
+  filtering: "none"
   filterWeight: 0.01
   filterCutoffRatio: 0.9
-  stressFormulation: 'no'
-  variableProperties: 'no'
+  stressFormulation: "no"
+  variableProperties: "no"
   p_residualTol: 1.0e-08
-  p_residualProj: 'no'
+  p_residualProj: "no"
   v_residualTol: 1.0e-08
-  v_residualProj: 'no'
+  v_residualProj: "no"
   density: 1.0
-  viscosity: -4600
-  advection: 'yes'
-  PARFWRITE: 'no'
-  PARFNAME: outparfile
+  viscosity: -3300 # Higher Reb
+  advection: "yes"
+
+  # Runtime parameters
+  PARFWRITE: "no"
+  PARFNAME: "outparfile"
   LOGLEVEL: 1
-  WALLTIME: '23:45'
-  READCHKPT: 'yes'
+  WALLTIME: "23:45"
+  READCHKPT: "yes"
   CHKPFNUMBER: 1
   CHKPINTERVAL: 10000000
   AVSTEP: 10
-  IOSTEP: 10000
-  SMPSTEP: 25
-runner:
-  RL_algorithm: TD3
-  nb_interactions: 2510
-  nb_episodes: 200
-  nb_warmup_episodes: 1
-  train_steps: 300
-  ckpt_int: 1
-  custom_policy: true
-  policy_file: ../conf/default_ddpg_policy.yml
-  learning_rate: 0.001
-  batch_size: 2048
-  n_epochs: 25
-  buffer_size: 20000000
-  rescale_actions: true
-  target_kl: 0.02
-  likhood_clipping: 0.5
-  gamma: 0.99
-  ent_coef: 0.0
-  max_grad_norm: 0.5
-  vf_coef: 0.5
-  action_noise: 0.1
-  gradient_steps: 64
-  size_history: 10
-  tau: 0.005
-  learning_starts: 100
-  policy_delay: 1
-  target_policy_noise: 0.2
-  target_noise_clip: 0.5
-  seed: -5000
-  custom_buffer: false
-  keep_frac: 0.25
-  buffer_mode: rotate
-  random_init: -1
-  agent_run_name: 401031
-  load_agent: true
-  rewrite_input_files: false
-  evaluation: true
-  rank: 6
-  policy: rl_model_265658400_steps
-  learnt_policy: true
-  vars_record: true
-  vars_record_freq: 1
-  rew_mode: Homo
-  normalize_input: utau
-  ctrl_min_amp: -0.061
-  ctrl_max_amp: 0.061
-  ctrl_array_size: 1
-  npl_state: 2
-  u_tau: 0.061
-  dUdy: 16.9
-logging:
-  run_name: 401031
-  group: null
-  notes: null
-  save_dir: ../runs
+  # During training we do not need write running statistics and time-series data
+  IOSTEP: 300000000
+  SMPSTEP: 1000