array_num_stps_eps = {}
array_re_all = np.zeros((n_agents,times,num_stp_int))
array_re_max = np.zeros((n_agents,times,num_stp_int))
for i in range(n_agents):
array_re[i] = np.mean([Q_list[i,m] for m in range(times)], axis=0)
array_num_stps_eps[i] = np.mean([Num_stps_per_eps[i,m] for m in range(times)], axis=0)
array_num_stps_eps[i] = np.repeat(np.arange(len(array_num_stps_eps[i])), array_num_stps_eps[i].astype(int))
with open('results/'+str(run)+'/array_re_'+str(i)+'.pkl', 'wb') as f:
pickle.dump(array_re[i], f,protocol=2)
with open('results/'+str(run)+'/array_num_stps_eps_'+str(i)+'.pkl', 'wb') as f:
pickle.dump(array_num_stps_eps[i], f,protocol=2)
for j in range(times):
ls = Q_list[i,j]
print(i,j)
array_re_all[i,j,:] = relative_error_plot(NUM_EPISODES_def, Qstar, ls)[0]
array[i,j,:]=plot_performance_curves(NUM_EPISODES_def,env, ls)[0]
with open('results/'+str(run)+'/array.pkl', 'wb') as f:
pickle.dump(array, f,protocol=2)
array_mean = np.zeros((n_agents,num_stp_int))
for i in range(n_agents):
array_mean[i]= np.mean(array[i], axis=0)
with open('results/'+str(run)+'/array_mean_'+str(i)+'.pkl', 'wb') as f:
pickle.dump(array_mean[i], f,protocol=2)
list_qstar = [performance_plot(env, Qstar)]*num_stp_int
plots('results',[array_mean[0], array_mean[1], list_qstar, array_mean[2], array_mean[3],array_mean[4]],
['LBQL','QL','OP', 'Double-QL','SQL', 'BCQL'],
['r','b', 'y', 'g', 'c','m'],
'Episodes',
f,
protocol=2)
else:
Q_list[i, j], elapsed_time[i, j] = agent.train()
with open(
'results/' + str(run) + '/' + str(i) + '_' + str(j) +
'.pkl', 'wb') as f:
pickle.dump([Q_list[i, j], elapsed_time[i, j]], f, protocol=2)
num_stp_int = int(np.ceil(NUM_STEPS_def / at))
array = np.zeros((n_agents, times, num_stp_int))
for i in range(n_agents):
for j in range(times):
ls = Q_list[i, j]
print(i, j)
array[i, j, :] = plot_performance_curves(at, NUM_STEPS_def, env, ls)[0]
with open('results/' + str(run) + '/array.pkl', 'wb') as f:
pickle.dump(array, f, protocol=2)
fig1 = plt.figure()
Epoch = np.tile(np.arange(0, array.shape[2]),
(array.shape[0] * array.shape[1]))
Alg = np.repeat(['LBQL', 'QL', 'SQL', 'Double-QL', 'BCQL'],
array.shape[1] * array.shape[2])
Returns = array.ravel()
df = pd.DataFrame()
df = pd.DataFrame(columns=['Epoch', 'Algo', 'Reward'])
df['Epoch'] = Epoch
df['Algo'] = Alg
df['Reward'] = Returns