def main():
TD_path = COMMON_PATH + 'TD/Data/' + str(lam) + '/'
ETD_path = COMMON_PATH + 'ETD/Data/' + str(lam) + '/'
# result_TD is a list of lists where each component list is
# [step_size, mean, SEM]
result_TD = get_mean_and_SEM_all_stepSizes(TD_path, curve,
performance_measure)
result_ETD = get_mean_and_SEM_all_stepSizes(ETD_path, curve,
performance_measure)
# Sort the results of both method according to the mean performance measure
result_TD.sort(key=lambda x: x[1])
result_ETD.sort(key=lambda x: x[1])
best_TD_step_size = result_TD[0][0]
best_TD_step_size_2RaisedTo = int(math.log(best_TD_step_size, 2))
best_ETD_step_size = result_ETD[0][0]
best_ETD_step_size_2RaisedTo = int(math.log(best_ETD_step_size, 2))
# Now that we have the best step size, let's go and plot the learning curve of that step size
best_TD_directory = COMMON_PATH + 'TD/Data/' + str(lam) + '/2' + str(
best_TD_step_size_2RaisedTo) + '/'
best_ETD_directory = COMMON_PATH + 'ETD/Data/' + str(lam) + '/2' + str(
best_ETD_step_size_2RaisedTo) + '/'
# measures_TD is a list of lists, where each element (each list)
# is the measures_TD of a trial. See performance_measure() for an example
measures_TD = get_all_files(best_TD_directory, performance_measure)
measures_ETD = get_all_files(best_ETD_directory, performance_measure)
# Plot the graph
mean_TD = np.mean(measures_TD, axis=0)
yerror_TD = np.std(measures_TD, axis=0) / math.sqrt(
len(measures_TD)
) # Approximate the standard error of the mean using std(X)/sqrt(n), where n = len(X)
mean_ETD = np.mean(measures_ETD, axis=0)
yerror_ETD = np.std(measures_ETD, axis=0) / math.sqrt(len(measures_ETD))
# plt.yscale('log')
plt.errorbar(np.arange(mean_TD.size),
mean_TD,
yerr=yerror_TD,
label=r'TD, $\alpha$=' + str(best_TD_step_size))
plt.errorbar(np.arange(mean_ETD.size),
mean_ETD,
yerr=yerror_ETD,
label=r'ETD, $\alpha$=' + str(best_ETD_step_size))
plt.xlabel('Episodes')
plt.ylabel('negated ' + str(performance_measure) +
' per episode\nAveraged over ' + str(len(measures_TD)) +
' runs',
rotation=0)
plt.title('Best learning curve for TD and ETD PuddleWorld control \n' +
r'$\lambda$ = ' + str(lam) + '\n' + str(curve_verbose))
plt.legend(loc=0)
plt.show()
def main():
# measures is a list of lists, where each element (each list)
# is the measures of a trial
measures = get_all_files(directory)
# Plot the graph
mean = np.mean(measures, axis=0)
plt.plot(np.arange(mean.size), mean, label=r'$\lambda$' + '=' + str(lam) + ', alpha=' + str(alpha))
plt.xlabel('Episodes')
plt.ylabel('Steps per Episode\nAveraged over ' + str(len(measures)) + ' runs', rotation=0)
plt.title('Learning Curve for ' + method + ' Control on MountainCar')
plt.legend(loc=0)
plt.show()
def main():
TD_path = COMMON_PATH + 'TD/Data/' + str(lam) + '/'
ETD_path = COMMON_PATH + 'ETD/Data/' + str(lam) + '/'
# result_TD is a list of lists where each component list is
# [step_size, mean, SEM]
result_TD = get_mean_and_SEM_all_stepSizes(TD_path, curve)
result_ETD = get_mean_and_SEM_all_stepSizes(ETD_path, curve)
# Sort the results of both method according to the mean performance measure
result_TD.sort(key=lambda x: x[1])
result_ETD.sort(key=lambda x: x[1])
best_TD_step_size = result_TD[0][0]
best_TD_step_size_2RaisedTo = int(math.log(best_TD_step_size, 2))
best_ETD_step_size = result_ETD[0][0]
best_ETD_step_size_2RaisedTo = int(math.log(best_ETD_step_size, 2))
# Now that we have the best step size, let's go and plot the learning curve of that step size
best_TD_directory = COMMON_PATH + 'TD/Data/' + str(lam) + '/2' + str(
best_TD_step_size_2RaisedTo) + '/'
best_ETD_directory = COMMON_PATH + 'ETD/Data/' + str(lam) + '/2' + str(
best_ETD_step_size_2RaisedTo) + '/'
# measures_TD is a list of lists, where each element (each list)
# is the measures_TD of a trial.
measures_TD = get_all_files(best_TD_directory)
measures_ETD = get_all_files(best_ETD_directory)
# Plot the graph
mean_TD = np.mean(measures_TD, axis=0)
yerror_TD = np.std(measures_TD, axis=0) / math.sqrt(
len(measures_TD)
) # Approximate the standard error of the mean using std(X)/sqrt(n), where n = len(X)
mean_ETD = np.mean(measures_ETD, axis=0)
yerror_ETD = np.std(measures_ETD, axis=0) / math.sqrt(len(measures_ETD))
fig, ax = plt.subplots(nrows=1, ncols=1)
plt.xticks([0, 10000, 20000, 30000], [0, '10K', '20K', '30K'])
ax.errorbar(np.arange(mean_TD.size),
mean_TD,
yerr=yerror_TD,
label=r'TD, $\alpha$=' + str(best_TD_step_size),
color='tab:blue')
ax.errorbar(np.arange(mean_ETD.size),
mean_ETD,
yerr=yerror_ETD,
label=r'ETD, $\alpha$=' + str(best_ETD_step_size),
color='tab:red')
ax.set_xlim(left=0, right=30000)
ax.set_ylim(bottom=20, top=30)
ax.set_xlabel('Episodes', fontsize=35)
ax.set_ylabel(r'$\sqrt{\widehat{\overline{VE}}}$',
rotation=0,
fontsize=35,
labelpad=45)
ax.tick_params(labelsize=30, which='major', axis='both')
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
fig.savefig('/Users/rlai/Desktop/test.pdf',
format='pdf',
dpi=300,
bbox_inches='tight')
# plt.title('Best learning curve for TD and ETD Prediction on MountainCar \n' + r'$\lambda$ = ' + str(lam) + '\n' + str(curve_verbose) )
# plt.title(r'$\lambda$ = ' + str(lam), fontsize=25)
# plt.legend(loc=0, fontsize=15)
plt.show()
plt.close(fig)