def main(): from arguments import get_args args = get_args() device = torch.device("cpu") pool = Pool() experiment_save_dir = get_experiment_save_dir(args) if args.debug: rollout(args, D_IN, D_OUT, pool, device, exp_save_dir=experiment_save_dir, pop_size=5, elite_prop=0.2) else: print('there should be only one of me!') rollout(args, D_IN, D_OUT, pool, device, exp_save_dir=experiment_save_dir, pop_size=args.pop_size)
def main(): args = get_args() print("start the train function") args.init_sigma = 0.6 args.lr = 0.001 # plot_weight_histogram(parameters) exp_save_dir = get_experiment_save_dir(args) inner_loop_ppo(args, args.lr, num_steps=1000, num_updates=4000, inst_on=False, visualize=False, save_dir=exp_save_dir)
if __name__ == "__main__": pop_size = 504 num_steps = 1500 args = get_args() # set up the parallelization try: from mpipool import Pool pool = Pool() except: pool = None experiment_save_dir = get_experiment_save_dir(args) env_name = register_set_goal(0) init_sigma = args.init_sigma envs = make_vec_envs(env_name, args.seed, 1, args.gamma, None, torch.device("cpu"), False) if args.load_ga: last_iter = get_start_gen_idx(args.load_ga, experiment_save_dir) - 1 start_weights = torch.load( os.path.join(experiment_save_dir, f"saved_weights_gen_{last_iter}.dat")) else: blueprint_model = init_ppo(envs, log(init_sigma)) start_weights = get_model_weights(blueprint_model) start_weights.append(np.array([args.lr]))