def settingSE2():
T = 100
hp1 = np.array([10, 1, 0.1, 0.01, 0.001, 0.0001])
hp2 = np.array([10, 1, 0.1, 0.01, 0.001, 0.0001])
print("Setting 2 SE started at " + str(datetime.datetime.now()),
flush=True)
exp = Experiment()
exp.initialize(timesteps=T, n_runs=5)
exp.add_problem('ExperimentSetting2-v0', {}, name='II')
val = [8, 16, 32, 64]
for p in val:
print("setting 2 SE p = " + str(p) + " started.", flush=True)
for lr in hp1:
param = {'lr': lr}
exp.add_method('RealAR', {
'p': p,
'optimizer': RealOGD(hyperparameters=param)
},
name='OGD_{0}'.format(lr))
for eta in hp1:
for eps in hp2:
param = {'eps': eps, 'eta': eta}
exp.add_method('RealAR', {
'p': p,
'optimizer': RealONS(hyperparameters=param)
},
name='ONS_{0}_{1}'.format(eta, eps))
print("Setting 2 SE tuned at " + str(datetime.datetime.now()), flush=True)
exp.scoreboard(n_digits=10)
def settingSE8():
d = 2
n = 20
T = 10000
print("Setting 8 SE started at " + str(datetime.datetime.now()),
flush=True)
exp = Experiment()
exp.initialize(timesteps=T, n_runs=n)
#problem
exp.add_problem('ExperimentSetting8-v0', name='VIII')
exp.add_method('RealAR', {
'p': 16,
'optimizer': RealOGD(hyperparameters={'lr': 10.0})
},
name='ARMA-OGD')
exp.add_method(
'RealAR', {
'p': 64,
'optimizer': RealONS(hyperparameters={
'eps': 0.1,
'eta': 0.1
})
},
name='ARMA-ONS')
exp.add_method('ArimaAR', {
'p': 16,
'd': d,
'optimizer': RealOGD(hyperparameters={'lr': 10.0})
},
name='ARIMA-OGD')
exp.add_method(
'ArimaAR', {
'p': 16,
'd': d,
'optimizer': RealONS(hyperparameters={
'eps': 10.0,
'eta': 1.0
})
},
name='ARIMA-ONS')
exp.add_method('HedgeAR', {'p': 16, 'c': -1}, name='Hedge-FTRL')
exp.add_method('HedgeARSE', {'p': 16, 'c': -1}, name='Hedge-ONS')
print("Setting 8 SE finished at " + str(datetime.datetime.now()),
flush=True)
exp.scoreboard(n_digits=10)
exp.graph(save_as=store_directory + "papersetting8SE.pdf",
avg_regret=True,
size=15,
start_time=100,
dpi=100,
save_csv_path=store_directory + 'papersetting8SE')
def settingAE8():
d = 1
n = 5
T = 100
print("Setting ae CarReg started at " + str(datetime.datetime.now()),
flush=True)
exp = Experiment()
exp.initialize(metrics=['ae'], timesteps=T, n_runs=n)
#problem
exp.add_problem('ExperimentSetting8-v0', name='VIII')
val = [8, 16, 32, 64]
hp = [10, 1, 0.1, 0.01, 0.001, 0.0001]
for p in val:
for lr in hp:
exp.add_method(
'RealAR', {
'p': p,
'optimizer': RealOGD(loss=ae, hyperparameters={'lr': lr})
},
name='ArmaOGD_{0:2d}_{1:2.4f}'.format(p, lr))
exp.add_method(
'ArimaAR', {
'p': p,
'd': 2,
'optimizer': RealOGD(loss=ae, hyperparameters={'lr': lr})
},
name='ArimaOGD_{0:2d}_{1:2.4f}'.format(p, lr))
print("Setting ae CarReg tuned at " + str(datetime.datetime.now()),
flush=True)
exp.scoreboard(n_digits=10, metric='ae')
def settingAE3():
ar_dyn = np.array([0.6, -0.5, 0.4, -0.4, 0.3])
ma_dyn = np.array([0.3, -0.2])
mean, noise_magnitude = 0, 0.3
p = 10
d = 1
n = 5
T = 100
print("Setting ae 3 started at " + str(datetime.datetime.now()),
flush=True)
exp = Experiment()
exp.initialize(metrics=['ae'], timesteps=T, n_runs=n)
#problem
exp.add_problem('ExperimentSetting3-v0', {}, name='III')
val = [8, 16, 32, 64]
hp = [10, 1, 0.1, 0.01, 0.001, 0.0001]
for p in val:
for lr in hp:
exp.add_method(
'RealAR', {
'p': p,
'optimizer': RealOGD(loss=ae, hyperparameters={'lr': lr})
},
name='ArmaOGD_{0:2d}_{1:2.4f}'.format(p, lr))
print("Setting ae 3 tuned at " + str(datetime.datetime.now()), flush=True)
exp.scoreboard(n_digits=10, metric='ae')
def settingAE4():
ar_dyn = np.array([0.6, -0.5, 0.4, -0.4, 0.3])
ma_dyn = np.array([0.3, -0.2])
mean, noise_magnitude = 0, 0.3
d = 2
n = 20
T = 10000
print("Setting ae 4 started at " + str(datetime.datetime.now()),
flush=True)
exp = Experiment()
exp.initialize(metrics=['ae'], timesteps=T, n_runs=n)
#problem
exp.add_problem('ExperimentSetting4-v0', {
'p': ar_dyn,
'q': ma_dyn,
'c': mean,
'noise_magnitude': noise_magnitude
},
name='IV')
exp.add_method('RealAR', {
'p': 16,
'optimizer': RealOGD(loss=ae, hyperparameters={'lr': 0.1})
},
name='ARMA-OGD')
exp.add_method('ArimaAR', {
'p': 16,
'd': d,
'optimizer': RealOGD(loss=ae, hyperparameters={'lr': 1.0})
},
name='ARIMA-OGD')
exp.add_method('HedgeAR', {
'p': 16,
'c': -1,
'loss': ae
},
name='Hedge-FTRL')
print("Setting ae 4 finished at " + str(datetime.datetime.now()),
flush=True)
exp.scoreboard(n_digits=10, metric='ae')
exp.graph(save_as=store_directory + "papersetting4_AE.pdf",
avg_regret=True,
size=15,
start_time=100,
metric='ae',
dpi=100,
save_csv_path=store_directory + 'papersetting4_AE')
def settingSE6():
T = 100
print("Setting SP500 SE started at " + str(datetime.datetime.now()),
flush=True)
exp = Experiment()
exp.initialize(timesteps=T, n_runs=5)
exp.add_problem('ExperimentSetting6-v0', name='VI')
val = [8, 16, 32, 64]
hp = [10, 1, 0.1, 0.01, 0.001, 0.0001]
for p in val:
print("setting SP500 SE p = " + str(p) + " started.", flush=True)
for lr in hp:
exp.add_method('AutoRegressor', {
'p': p,
'optimizer': RealOGD(hyperparameters={'lr': lr})
},
name='ArmaOGD_{0:2d}_{1:2.4f}'.format(p, lr))
exp.add_method('ArimaAR', {
'p': p,
'd': 2,
'optimizer': RealOGD(hyperparameters={'lr': lr})
},
name='ArimaOGD_{0:2d}_{1:2.4f}'.format(p, lr))
for eps in hp:
for eta in hp:
exp.add_method('AutoRegressor', {
'p':
p,
'optimizer':
RealONS(hyperparameters={
'eps': eps,
'eta': eta
})
},
name='ArmaONS_{0:2d}_{1:2.4f}_{2:2.4f}'.format(
p, eps, eta))
exp.add_method('ArimaAR', {
'p':
p,
'd':
2,
'optimizer':
RealONS(hyperparameters={
'eps': eps,
'eta': eta
})
},
name='ArimaONS_{0:2d}_{1:2.4f}_{2:2.4f}'.format(
p, eps, eta))
print("Setting SP500 SE tuned at " + str(datetime.datetime.now()),
flush=True)
exp.scoreboard(n_digits=10)
def settingSE1():
ar_dyn = np.array([0.6, -0.5, 0.4, -0.4, 0.3])
ma_dyn = np.array([0.3, -0.2])
mean, noise_magnitude = 0, 0.3
n = 20
T = 10000
print("Setting 1 SE started at " + str(datetime.datetime.now()),
flush=True)
exp = Experiment()
exp.initialize(timesteps=T, n_runs=n)
#problem
exp.add_problem('ARMA-v0', {
'p': ar_dyn,
'q': ma_dyn,
'c': mean,
'noise_magnitude': noise_magnitude
},
name='I')
exp.add_method(
'RealAR', {
'p': 8,
'optimizer': RealONS(hyperparameters={
'eps': 1.0,
'eta': 1.0
})
},
name='ARMA-ONS')
exp.add_method('RealAR', {
'p': 8,
'optimizer': RealOGD(hyperparameters={'lr': 10})
},
name='ARMA-OGD')
exp.add_method('HedgeAR', {'p': 16, 'c': -1}, name='Hedge-FTRL')
exp.add_method('HedgeARSE', {'p': 16, 'c': -1}, name='Hedge-ONS')
print("Setting 1 SE finished at " + str(datetime.datetime.now()),
flush=True)
exp.scoreboard(n_digits=10)
exp.graph(save_as=store_directory + "papersetting1SE.pdf",
avg_regret=True,
size=15,
start_time=100,
dpi=100,
save_csv_path=store_directory + 'papersetting1SE')
def settingAE8():
d = 2
n = 20
T = 10000
print("Setting ae 8 started at " + str(datetime.datetime.now()),
flush=True)
exp = Experiment()
exp.initialize(metrics=['ae'], timesteps=T, n_runs=n)
#problem
exp.add_problem('ExperimentSetting8-v0', name='VIII')
exp.add_method('RealAR', {
'p': 8,
'optimizer': RealOGD(loss=ae, hyperparameters={'lr': 10.0})
},
name='ARMA-OGD')
exp.add_method('ArimaAR', {
'p': 16,
'd': d,
'optimizer': RealOGD(loss=ae, hyperparameters={'lr': 0.1})
},
name='ARIMA-OGD')
exp.add_method('HedgeAR', {
'p': 16,
'c': -1,
'loss': ae
},
name='Hedge-FTRL')
print("Setting ae 8 finished at " + str(datetime.datetime.now()),
flush=True)
exp.scoreboard(n_digits=10, metric='ae')
exp.graph(save_as=store_directory + "papersetting8_AE.pdf",
avg_regret=True,
size=15,
start_time=100,
dpi=100,
metric='ae',
save_csv_path=store_directory + "papersetting8_MA")
def settingSE1():
T = 100
p = 10
ar_dyn = np.array([0.6, -0.5, 0.4, -0.4, 0.3])
ma_dyn = np.array([0.3, -0.2])
mean, noise_magnitude = 0, 0.3
hp1 = np.array([10, 1, 0.1, 0.01, 0.001, 0.0001])
hp2 = np.array([10, 1, 0.1, 0.01, 0.001, 0.0001])
print("Setting 1 SE started at " + str(datetime.datetime.now()),
flush=True)
exp = Experiment()
exp.initialize(timesteps=T, n_runs=5)
exp.add_problem('ARMA-v0', {
'p': ar_dyn,
'q': ma_dyn,
'c': mean,
'noise_magnitude': noise_magnitude
},
name='I')
val = [8, 16, 32, 64]
for p in val:
print("setting 1 SE p = " + str(p) + " started.", flush=True)
for lr in hp1:
param = {'lr': lr}
exp.add_method('RealAR', {
'p': p,
'optimizer': RealOGD(hyperparameters=param)
},
name='OGD_{0}'.format(lr))
for eta in hp1:
for eps in hp2:
param = {'eps': eps, 'eta': eta}
exp.add_method('RealAR', {
'p': p,
'optimizer': RealONS(hyperparameters=param)
},
name='ONS_{0}_{1}'.format(eta, eps))
print("Setting 1 SE tuned at " + str(datetime.datetime.now()), flush=True)
exp.scoreboard(n_digits=10)