def parallel_train_pipeline(config, methods, env, eval_qnet, bhv_qnet, seedvec, max_name_length):
num_method = len(methods)
mse = np.zeros(len(methods))
ind_mse = np.zeros(len(methods))
mse_w = np.zeros(len(methods))
results, target = train_pipeline(env, config, eval_qnet, bhv_qnet, seedvec)
for i_method in range(num_method):
mse_1, mse_2, mse_3 = error_info(results[i_method], target, methods[i_method].ljust(max_name_length))
mse[i_method] = mse_1
ind_mse[i_method] = mse_2
mse_w[i_method] = mse_3
return(mse, ind_mse, mse_w)
env = gym.make("ContinuousCartPole-v0")
config = config.contcartpole_config
agent = load_ddpg_agent(config,
filename=args.env_name + '_ddpg_agent.pth.tar')
agent.actor.eval()
agent.critic.eval()
seedvec = np.random.randint(0, config.MAX_SEED, config.sample_num_traj)
factual_types = ['hard', 2.5, 2.0, 1.5, 1.0, 0.5]
methods = ['Baseline'] + ['mse_pi_{}'.format(ft) for ft in factual_types] +\
['repbm_{}'.format(ft) for ft in factual_types]
num_method = len(methods)
max_name_length = len(max(methods, key=len))
mse = []
ind_mse = []
results, target = train_pipeline(env, config, agent, factual_types,
seedvec)
for i_method in range(num_method):
mse_1, mse_2 = error_info(results[i_method], target,
methods[i_method].ljust(max_name_length))
mse.append(mse_1)
ind_mse.append(mse_2)
print(mse)
print(ind_mse)
np.save('results/result_{}_{}'.format(args.env_name, args.pid),
[mse, ind_mse])