import os import socket import numpy as np from experiment_launcher import Launcher from experiment_launcher.utils import is_local ######################################################################################################################## # EXPERIMENT PARAMETERS SETUP # SELECT ONE env_id, robot_id, num_contexts, num_trajectories_per_context, threshold_start_goal_pos, obstacle_cutoff_margin = 'EnvSimple2D', 'RobotPointMass', 500, 20, 1, 0.02 # env_id, robot_id, num_contexts, num_trajectories_per_context, threshold_start_goal_pos, obstacle_cutoff_margin = 'EnvNarrowPassageDense2D', 'RobotPointMass', 500, 20, 1, 0.02 # env_id, robot_id, num_contexts, num_trajectories_per_context, threshold_start_goal_pos, obstacle_cutoff_margin = 'EnvDense2D', 'RobotPointMass', 500, 20, 1, 0.02 # env_id, robot_id, num_contexts, num_trajectories_per_context, threshold_start_goal_pos, obstacle_cutoff_margin = 'EnvSpheres3D', 'RobotPanda', 500, 20, 1.83, 0.05 # 1.83 = 7 * np.deg2rad(15) ######################################################################################################################## # LAUNCHER hostname = socket.gethostname() LOCAL = is_local() TEST = False # USE_CUDA = True USE_CUDA = False N_SEEDS = num_contexts N_EXPS_IN_PARALLEL = 15 if not USE_CUDA else 1 # N_CORES = N_EXPS_IN_PARALLEL N_CORES = 8 MEMORY_SINGLE_JOB = 12000 MEMORY_PER_CORE = N_EXPS_IN_PARALLEL * MEMORY_SINGLE_JOB // N_CORES PARTITION = 'gpu' if USE_CUDA else 'amd3,amd2,amd' GRES = 'gpu:1' if USE_CUDA else None # gpu:rtx2080:1, gpu:rtx3080:1, gpu:rtx3090:1, gpu:a5000:1 CONDA_ENV = 'mpd-public' exp_name = f'generate_trajectories' os.environ["HDF5_USE_FILE_LOCKING"] = "FALSE" launcher = Launcher( exp_name=exp_name, exp_file='generate_trajectories', # project_name='project01234', n_seeds=N_SEEDS, n_exps_in_parallel=N_EXPS_IN_PARALLEL, n_cores=N_CORES, memory_per_core=MEMORY_PER_CORE, days=0, hours=7, minutes=59, seconds=0, partition=PARTITION, conda_env=CONDA_ENV, gres=GRES, use_timestamp=True ) ######################################################################################################################## # RUN launcher.add_experiment( env_id__=env_id, robot_id__=robot_id, num_trajectories=num_trajectories_per_context, threshold_start_goal_pos=threshold_start_goal_pos, obstacle_cutoff_margin=obstacle_cutoff_margin, device='cuda' if USE_CUDA else 'cpu', debug=False ) launcher.run(LOCAL, TEST)