def exploit_nstep(params) -> None:
path = os.getcwd() + "/data/save"
steps = [1, 5, 10, 15, 20]
for j in ['policy_loss', 'critic_loss', 'reward', 'duration']:
mean_list = []
std_list = []
for i in steps:
mean, std = plot_data(
path + '/' + j + '_nstep_' + str(i) + '_' + params.env_name +
'.txt', j + '_nstep_' + str(i))
mean_list.append(mean[-1])
std_list.append(std[-1])
plt.title(params.env_name)
plt.xlabel("Episodes")
plt.ylabel(j)
plt.legend(loc="lower right")
plt.savefig(path + '/../results/' + j + '_nstep_' +
make_full_string(params) + '.pdf')
plt.show()
plt.plot(steps, mean_list, label="bias")
plt.plot(steps, std_list, label="variance")
plt.title(params.env_name)
plt.xlabel("N in N-step")
plt.ylabel('variance, bias')
plt.legend(loc="lower right")
plt.savefig(path + '/../results/bias_variance_' + j + '_' +
make_full_string(params) + '.pdf')
plt.show()
def exploit_nstep_diff(params) -> None:
path = os.getcwd() + "/data/save"
steps = [1, 5, 10, 20]
mean_list = []
std_list = []
for i in steps:
mean, std = plot_data(
path + '/diff_' + str(i) + '_' + params.env_name + '.txt',
'nstep_' + str(i))
# print('n:', i, ' mean :', mean[-1], ' std:', std[-1])
mean_list.append(mean[-1])
std_list.append(std[-1])
plt.title(params.env_name)
plt.xlabel("Episodes")
plt.ylabel("diff")
plt.legend(loc="lower right")
plt.savefig(path + '/../results/diff_nstep_' + make_full_string(params) +
'.pdf')
plt.show()
plt.plot(steps, mean_list, label="bias")
plt.plot(steps, std_list, label="variance")
plt.title(params.env_name)
plt.xlabel("N in N-step")
plt.ylabel('variance, bias')
plt.legend(loc="lower right") # , bbox_to_anchor=(1, 0.5)
plt.savefig(path + '/../results/bias_variance_' +
make_full_string(params) + '.pdf')
plt.show()
def plot_beta_results(params) -> None:
path = os.getcwd() + "/data/save"
for beta in [0.1, 0.5, 1.0, 5.0, 10.0]:
plot_data(path + "/reward_" + str(beta) + '_' + params.env_name + '.txt', "reward " + str(beta))
plt.title(params.env_name)
plt.xlabel("Episodes")
plt.ylabel("Reward")
plt.legend(loc="lower right")
plt.savefig(path + '/../results/rewards_' + make_full_string(params) + '.pdf')
plt.show()
def exploit_cov_full(params) -> None:
path = os.getcwd() + "/data/save"
study = params.gradients
for i in range(len(study)):
plot_data(path + "/covariance_" + '.txt', "cem covariance norm ")
plt.title(params.env_name)
plt.xlabel("Episodes")
plt.ylabel("Covariance Norm")
plt.legend(loc="best")
plt.savefig(path + '/../results/covariance_' + make_full_string(params) + '.pdf')
plt.clf()
def exploit_reward_full(params) -> None:
path = os.getcwd() + "/data/save"
study = params.gradients
for i in range(len(study)):
plot_data(path + "/reward_" + params.study_name+study[i] + '_' + params.env_name + '.txt', params.study_name +'_'+"reward " + study[i])
plt.title(params.env_name)
plt.xlabel("Episodes")
plt.ylabel("Reward")
plt.legend(loc="best")
plt.savefig(path + '/../results/rewards_' + make_full_string(params) + '.pdf')
plt.show()
def exploit_duration_full(params) -> None:
path = os.getcwd() + "/data/save"
study = params.gradients
for i in range(len(study)):
plot_data(path + "/duration_" + study[i] + '_' + params.env_name + '.txt', "duration " + study[i])
plt.xlabel("Episodes")
plt.ylabel("Duration")
plt.legend(loc="lower right")
plt.title(params.env_name)
plt.savefig(path + '/../results/durations_' + make_full_string(params) + 'pg.pdf')
plt.show()
def exploit_policy_loss_full(params) -> None:
path = os.getcwd() + "/data/save"
study = params.gradients
for i in range(len(study)):
plot_data(path + "/policy_loss_" + study[i] + '_' + params.env_name + '.txt', "policy loss " + study[i])
plt.xlabel("Cycles")
plt.ylabel("Loss")
plt.legend(loc="lower right")
plt.title(params.env_name)
plt.savefig(path + '/../results/policy_loss_' + make_full_string(params) + 'pg.pdf')
plt.show()
def exploit_angles_global_full(params) -> None:
path = os.getcwd() + "/data/save"
study = params.gradients
for i in range(len(study)):
plot_data(path + "/angle_" + params.study_name+study[i] + '_' + params.env_name + '.txt', "scalar product " )
plt.title(params.env_name)
plt.xlabel("Episodes")
plt.ylabel("Scalar Product")
plt.legend(loc="best")
plt.savefig(path + '/../results/angles_' + make_full_string(params) + '.pdf')
plt.clf()
def exploit_distance_solo(params) -> None:
path = os.getcwd() + "/data/save"
study = params.gradients
for i in range(len(study)):
plot_data(path + "/reward_" + params.study_name+study[i] + '_' + params.env_name + '.txt', "distance")
# plot_data(path + "/reward_" + 'evo_pg'+study[i] + '_' + params.env_name + '.txt', "reward evo_pg")
plt.title(params.env_name)
plt.xlabel("Episodes")
plt.ylabel("distance")
plt.legend(loc="lower right")
plt.savefig(path + '/../results/distances_' + make_full_string(params) + '.pdf')
plt.clf()
def exploit_angles_global_full_comparison(params) -> None:
path = os.getcwd() + "/data/save"
study = params.gradients
for i in range(len(study)):
plot_data(path + "/angle_" + 'pg'+study[i] + '_' + params.env_name + '.txt', "scalar product pg")
plot_data(path + "/angle_" + 'cem'+study[i] + '_' + params.env_name + '.txt', "scalar product cem")
# plot_data(path + "/reward_" + 'evo_pg'+study[i] + '_' + params.env_name + '.txt', "reward evo_pg")
plt.title(params.env_name)
plt.xlabel("Episodes")
plt.ylabel("scalar product")
plt.legend(loc="lower right")
plt.savefig(path + '/../results/angles_' + make_full_string(params) + '.pdf')
plt.clf()
def exploit_critic_loss_full(params) -> None:
path = os.getcwd() + "/data/save"
study = params.gradients
for i in range(len(study)):
plot_data(
path + "/critic_loss_" + study[i] + '_' + params.env_name + '.txt',
"critic loss " + study[i])
plt.xticks(range(params.nb_cycles - 1), labels=arange(1, params.nb_cycles))
plt.xlabel("Cycles")
plt.ylabel("Loss")
plt.legend(loc="upper right")
plt.title(params.env_name)
plt.savefig(path + '/../results/critic_loss_' + make_full_string(params) +
'pg.pdf')
plt.show()
def exploit_reward_full(params) -> None:
path = os.getcwd() + "/data/save"
study = params.gradients
for i in range(len(study)):
plot_data(
path + "/reward_" + params.experiment + study[i] + '_' +
params.env_name + '.txt', "reward " + params.experiment)
if params.eval_freq != 1:
plt.xticks(range((params.nb_cycles - 1) // params.eval_freq + 1),
labels=arange(0, params.nb_cycles,
params.eval_freq).astype(str))
plt.title(params.env_name)
plt.xlabel("Episodes")
plt.ylabel("Reward")
plt.legend(loc="best")
plt.savefig(path + '/../results/rewards_' + make_full_string(params) +
'.pdf')
plt.show()
def exploit_angles_full(params) -> None:
path = os.getcwd() + "/data/save"
study = params.gradients
for i in range(len(study)):
for cycle in range(params.nb_cycles):
plot_data(path + "/gradient_angles_" +"#"+ str(cycle) +'.txt', params.study_name +'_'+"gradient_angles " )
plt.title(params.env_name)
plt.xlabel("traj_in_batch")
plt.ylabel("angle")
plt.legend(loc="best")
plt.savefig(path + '/../results/gradient_angles/gradient_angles_'+'#'+str(cycle) + make_full_string(params) + '.pdf')
plt.clf()
