def run(run_index,
model,
folder="data_dumps",
hp_opt="IS0",
sample_domain=1000):
'''
This function will run CO optimization using one of several coregionalization methods.
1. Pearson R Intrinsic Coregionalization Method (PRICM). This approach
will dynamically calculate the Pearson R value for the off-diagonals
in the ICM. Diagonals are kept as 1.0.
2. Intrinsic Coregionalization Method (ICM). This approach will use a
lower triangular matrix (L) of hyperparameters to generate the
coregionalization matrix B = LL^T.
Further, we can parameterize the hyperparameters in many ways:
1. IS0 - Only parameterize hyperparameters using values sampled at IS0.
2. Full - Parameterize hyperparameters using all sampled data.
3. Overlap - Parameterize hyperparameters using data that overlaps all IS.
**Parameters**
run_index: *int*
This is simply used for a naming convention.
model: *str*
The model to be used (PRICM or ICM).
folder: *str, optional*
What to name the folder where the data will go.
hp_opt: *str, optional*
With what data should the hyperparameters be parameterized.
Options: IS0, full, overlap
sample_domain: *int, optional*
How many data points to sample from the domain.
'''
hp_opt = hp_opt.lower()
allowed_hp_opt = ["is0", "full", "overlap"]
assert hp_opt in allowed_hp_opt, "Error, hp_opt (%s) not in %s" % (
hp_opt, ", ".join(allowed_hp_opt))
# Generate the main object
sim = Optimizer()
# Assign simulation properties
sim.hyperparameter_objective = MLE
###################################################################################################
# File names
sim.fname_out = None
sim.fname_historical = None
sim.logger_fname = "%s/%d_%s_%s.log" % (folder, run_index, model, hp_opt)
if os.path.exists(sim.logger_fname):
os.system("rm %s" % sim.logger_fname)
os.system("touch %s" % sim.logger_fname)
sim.obj_vs_cost_fname = None
sim.mu_fname = None
sim.sig_fname = None
sim.combos_fname = None
sim.hp_fname = None
sim.acquisition_fname = None
sim.save_extra_files = True
# Information sources, in order from expensive to cheap
rosenbrock = lambda x1, x2: (1.0 - x1)**2 + 100.0 * (x2 - x1**2)**2 - 456.3
sim.IS = [
lambda x1, x2: -1.0 * rosenbrock(x1, x2), lambda x1, x2: -1.0 *
(rosenbrock(x1, x2) + 10.0 * np.sin(10.0 * x1 + 5.0 * x2))
]
sim.costs = [1000.0, 1.0]
sim.save_extra_files = False
########################################
sim.numerical = True
sim.historical_nsample = 5
sim.domain = [(-2.0, 2.0), (-2.0, 2.0)]
sim.sample_n_from_domain = sample_domain
########################################
sim.n_start = 10 # The number of starting MLE samples
# sim.reopt = 20
sim.reopt = float('inf') # Never re-opt hyperparams
sim.ramp_opt = None
sim.parallel = False
# Parameters for debugging and overwritting
sim.debug = False
sim.verbose = True
sim.overwrite = True # If True, warning, else Error
sim.acquisition = getNextSample_misokg
# Functional forms of our mean and covariance
sim.mean = lambda X, Y, theta: np.array([-456.3 for _ in Y])
def cov_miso(X0, Y, theta, split=False):
Kx = squared(np.array(X0), [theta.l1], theta.sig_1)
Kx_l = squared(np.array(X0), [theta.l2], theta.sig_2)
return np.block([[Kx, Kx], [Kx, Kx + Kx_l]])
def cov_pricm(X0, Y, theta, split=False):
Kx = squared(np.array(X0), [theta.l1], theta.sig_1)
Kx = Kx + 1E-6 * np.eye(Kx.shape[0])
if model.lower() == "pricm":
Ks = np.array([
np.array([
theta.rho[str(sorted([i, j]))] for j in range(theta.n_IS)
]) for i in range(theta.n_IS)
])
elif model.lower() == "icm":
L = np.array([
np.array([
theta.rho[str(sorted([i, j]))] if i >= j else 0.0
for j in range(theta.n_IS)
]) for i in range(theta.n_IS)
])
# Force it to be positive semi-definite
Ks = L.dot(L.T)
if split:
return Ks, Kx
else:
return np.kron(Ks, Kx)
sim.theta.bounds = {}
sim.theta.sig_1, sim.theta.bounds['sig_1'] = None, (1E-2,
lambda _, Y: np.var(Y))
sim.theta.l1, sim.theta.bounds['l1'] = None, (1E-1, 1)
if model == "miso":
sim.cov = cov_miso
sim.theta.sig_2, sim.theta.bounds['sig_2'] = None, (
1E-2, lambda _, Y: np.var(Y))
sim.theta.l2, sim.theta.bounds['l2'] = None, (1E-1, 1)
sim.theta.rho = {
str(sorted([i, j])): 1.0
for i in range(len(sim.IS)) for j in range(i, len(sim.IS))
}
else:
sim.cov = cov_pricm
sim.theta.rho = {"[0, 0]": None, "[0, 1]": None, "[1, 1]": None}
if model.lower() == "icm":
sim.theta.rho = {
str(sorted([i, j])): None
for i in range(len(sim.IS)) for j in range(i, len(sim.IS))
}
elif model.lower() == "pricm":
sim.theta.rho = {
str(sorted([i, j])): 1.0
for i in range(len(sim.IS)) for j in range(i, len(sim.IS))
}
sim.dynamic_pc = True
else:
raise Exception("Invalid model. Use MISO, ICM, or PRICM")
for k in sim.theta.rho.keys():
sim.theta.bounds['rho %s' % k] = (0.1, 1.0)
a, b = eval(k)
if a != b:
sim.theta.bounds['rho %s' % k] = (0.01, 1.0 - 1E-6)
sim.theta.set_hp_names()
# Define how we update hyperparameters
hp_opt = hp_opt.lower()
if hp_opt == "is0":
sim.update_hp_only_with_IS0 = True
sim.update_hp_only_with_overlapped = False
elif hp_opt == "overlap":
sim.update_hp_only_with_IS0 = False
sim.update_hp_only_with_overlapped = True
elif hp_opt == "full":
sim.update_hp_only_with_IS0 = False
sim.update_hp_only_with_overlapped = False
else:
raise Exception("Unknown hp_opt (%s)." % hp_opt)
# These should be False by default, but ensure they are
sim.theta.normalize_L = False
sim.theta.normalize_Ks = False
sim.preconditioned = False
# Assign our likelihood function.
sim.loglike = g_loglike
###################################################################################################
# Start simulation
sim.iteration_kill_switch = None
sim.cost_kill_switch = 100000
sim.run()
def run(run_index, folder="data_dumps", sample_domain=1000):
'''
This function will run CO optimization using one of several coregionalization methods.
1. Pearson R Intrinsic Coregionalization Method (PRICM). This approach
will dynamically calculate the Pearson R value for the off-diagonals
in the ICM. Diagonals are kept as 1.0.
2. Intrinsic Coregionalization Method (ICM). This approach will use a
lower triangular matrix (L) of hyperparameters to generate the
coregionalization matrix B = LL^T.
Further, we can parameterize the hyperparameters in many ways:
1. IS0 - Only parameterize hyperparameters using values sampled at IS0.
2. Full - Parameterize hyperparameters using all sampled data.
3. Overlap - Parameterize hyperparameters using data that overlaps all IS.
**Parameters**
run_index: *int*
This is simply used for a naming convention.
folder: *str, optional*
What to name the folder where the data will go.
sample_domain: *int, optional*
How many data points to sample from the domain.
'''
# Generate the main object
sim = Optimizer()
# Assign simulation properties
sim.hyperparameter_objective = MLE
sim.acquisition = getNextSample_EI
###################################################################################################
# File names
sim.fname_out = None
sim.fname_historical = None
sim.logger_fname = "%s/%d_%s.log" % (folder, run_index, "ei")
if os.path.exists(sim.logger_fname):
os.system("rm %s" % sim.logger_fname)
os.system("touch %s" % sim.logger_fname)
sim.obj_vs_cost_fname = None
sim.mu_fname = None
sim.sig_fname = None
sim.combos_fname = None
sim.hp_fname = None
sim.acquisition_fname = None
sim.save_extra_files = True
# Information sources, in order from expensive to cheap
IS0 = pickle.load(open("IS0.pickle", 'r'))
sim.IS = [
lambda x1: -1.0 * IS0[int((x1 - 0.5) * 1000.0)][0],
]
sim.costs = [
np.mean([IS[1] for IS in IS0]),
]
sim.save_extra_files = False
########################################
sim.numerical = True
sim.historical_nsample = 5
sim.domain = [(0.5, 2.5)]
sim.sample_n_from_domain = sample_domain
########################################
sim.n_start = 10 # The number of starting MLE samples
# sim.reopt = 20
sim.reopt = float("inf")
sim.ramp_opt = None
sim.parallel = False
# Parameters for debugging and overwritting
sim.debug = False
sim.verbose = True
sim.overwrite = True # If True, warning, else Error
# Functional forms of our mean and covariance
sim.mean = lambda X, Y, theta: np.array([0.0 for _ in Y])
def cov(X0, Y, theta):
return squared(np.array(X0), [theta.l1], theta.sig_1)
sim.cov = cov
sim.theta.bounds = {}
sim.theta.sig_1, sim.theta.bounds['sig_1'] = None, (1E-2,
lambda _, Y: np.var(Y))
sim.theta.l1, sim.theta.bounds['l1'] = None, (1E-1, 1)
sim.theta.rho = {
str(sorted([i, j])): 1.0
for i in range(len(sim.IS)) for j in range(i, len(sim.IS))
}
for k in sim.theta.rho.keys():
sim.theta.bounds['rho %s' % k] = (0.1, 1.0)
a, b = eval(k)
if a != b:
sim.theta.bounds['rho %s' % k] = (0.01, 1.0 - 1E-6)
sim.theta.set_hp_names()
# Define how we update hyperparameters
sim.update_hp_only_with_IS0 = False
sim.update_hp_only_with_overlapped = False
# These should be False by default, but ensure they are
sim.theta.normalize_L = False
sim.theta.normalize_Ks = False
sim.preconditioned = False
# Assign our likelihood function.
sim.loglike = g_loglike
###################################################################################################
# Start simulation
sim.iteration_kill_switch = None
sim.cost_kill_switch = 3000
sim.run()
def run_misokg(run_index,
sffx="misokg",
scaled=False,
loose=False,
very_loose=False,
use_MAP=False,
upper=1.0,
run_unitary=False,
use_miso=False):
SAMPLE_DOMAIN = 1000
# Generate the main object
sim = Optimizer()
# Assign simulation properties
if use_MAP:
sim.hyperparameter_objective = MAP
else:
sim.hyperparameter_objective = MLE
###################################################################################################
# File names
sim.fname_out = None
sim.fname_historical = None
sim.logger_fname = "data_dumps/%d_%s.log" % (run_index, sffx)
if os.path.exists(sim.logger_fname):
os.system("rm %s" % sim.logger_fname)
os.system("touch %s" % sim.logger_fname)
sim.obj_vs_cost_fname = None
sim.mu_fname = None
sim.sig_fname = None
sim.combos_fname = None
#sim.hp_fname = None
sim.hp_fname = "data_dumps/%d_HP_%s.log" % (run_index, sffx)
sim.acquisition_fname = None
sim.save_extra_files = True
# Information sources, in order from expensive to cheap
rosenbrock = lambda x1, x2: (1.0 - x1)**2 + 100.0 * (x2 - x1**2)**2 - 456.3
sim.IS = [
lambda x1, x2: -1.0 * rosenbrock(x1, x2), lambda x1, x2: -1.0 *
(rosenbrock(x1, x2) + 0.1 * np.sin(10.0 * x1 + 5.0 * x2))
]
#sim.IS = [
# lambda x1, x2: (1.0 - x1)**2 + 100.0 * (x2 - x1**2)**2 - 456.3 + np.random.normal()
# lambda x1, x2: (1.0 - x1)**2 + 100.0 * (x2 - x1**2)**2 - 456.3 + 2.0 * np.sin(10.0 * x1 + 5.0 * x2)
#]
sim.costs = [1000.0, 1.0]
########################################
sim.numerical = True
sim.historical_nsample = 5
sim.domain = [(-2.0, 2.0), (-2.0, 2.0)]
sim.sample_n_from_domain = SAMPLE_DOMAIN
########################################
sim.n_start = 10 # The number of starting MLE samples
sim.reopt = 20
sim.ramp_opt = None
sim.parallel = False
# Parameters for debugging and overwritting
sim.debug = False
sim.verbose = True
sim.overwrite = True # If True, warning, else Error
sim.acquisition = getNextSample_misokg
# Functional forms of our mean and covariance
sim.mean = lambda X, Y, theta: np.array([-456.3 for _ in Y])
def cov_miso(X0, Y, theta):
Kx = squared(np.array(X0)[:, 1:], [theta.l1, theta.l2], theta.sig_1)
Kx_l = squared(np.array(X0)[:, 1:], [theta.l3, theta.l4], theta.sig_2)
return np.block([[Kx, Kx], [Kx, Kx + Kx_l]])
def cov_bonilla(X0, Y, theta):
#Kx = mk52(np.array(X0)[:, 1:], [theta.l1, theta.l2], theta.sig_m)
Kx = squared(np.array(X0)[:, 1:], [theta.l1, theta.l2], theta.sig_m)
Kx = Kx + 1E-6 * np.eye(Kx.shape[0])
if run_unitary:
Ks = np.array([[1.0, 1.0 - 1E-6], [1.0 - 1E-6, 1.0]])
else:
L = np.array([
np.array([
theta.rho[str(sorted([i, j]))] if i >= j else 0.0
for j in range(theta.n_IS)
]) # Lower triangulary
for i in range(theta.n_IS)
])
Ks = L.dot(L.T)
e = np.diag(np.array([theta.e1, theta.e2]))
Ks = np.matmul(e, np.matmul(Ks, e))
K = np.kron(Ks, Kx)
return np.kron(Ks, Kx)
if use_miso:
sim.cov = cov_miso
sim.theta.bounds = {}
sim.theta.sig_1, sim.theta.bounds['sig_1'] = None, (
1E-2, lambda _, Y: np.var(Y))
sim.theta.sig_2, sim.theta.bounds['sig_2'] = None, (
1E-2, lambda _, Y: np.var(Y))
sim.theta.l1, sim.theta.bounds['l1'] = None, (1E-1, 1)
sim.theta.l2, sim.theta.bounds['l2'] = None, (1E-1, 1)
sim.theta.l3, sim.theta.bounds['l3'] = None, (1E-1, 1)
sim.theta.l4, sim.theta.bounds['l4'] = None, (1E-1, 1)
sim.theta.rho = {"[0, 0]": 1.0, "[0, 1]": 1.0, "[1, 1]": 1.0}
sim.theta.bounds['rho [0, 0]'] = (1.0, 1.0)
sim.theta.bounds['rho [0, 1]'] = (1.0, 1.0)
sim.theta.bounds['rho [1, 1]'] = (1.0, 1.0)
else:
sim.cov = cov_bonilla
sim.theta.bounds = {}
sim.theta.sig_m, sim.theta.bounds['sig_m'] = None, (
1E-2, lambda _, Y: np.var(Y))
sim.theta.l1, sim.theta.bounds['l1'] = None, (1E-1, 1)
sim.theta.l2, sim.theta.bounds['l2'] = None, (1E-1, 1)
if scaled:
sim.theta.e1, sim.theta.bounds['e1'] = None, (1E-1, upper)
sim.theta.e2, sim.theta.bounds['e2'] = None, (1E-1, upper)
else:
sim.theta.e1, sim.theta.bounds['e1'] = 1.0, (1E-1, upper)
sim.theta.e2, sim.theta.bounds['e2'] = 1.0, (1E-1, upper)
# NOTE! This is a reserved keyword in misoKG. We will generate a list of the same length
# of the information sources, and use this for scaling our IS.
if very_loose:
sim.theta.rho = {"[0, 0]": None, "[0, 1]": None, "[1, 1]": None}
elif loose:
sim.theta.rho = {"[0, 0]": 1.0, "[0, 1]": None, "[1, 1]": 1.0}
elif run_unitary:
sim.theta.rho = {"[0, 0]": 1.0, "[0, 1]": 1.0, "[1, 1]": 1.0}
else:
raise Exception("What is trying to be run?")
sim.theta.bounds['rho [0, 0]'] = (0.1, upper)
sim.theta.bounds['rho [0, 1]'] = (0.1, upper)
sim.theta.bounds['rho [1, 1]'] = (0.01, upper)
sim.theta.set_hp_names()
sim.update_hp_only_with_IS0 = False
sim.update_hp_only_with_overlapped = False
sim.theta.normalize_L = False
sim.theta.normalize_Ks = False
###################################################################################################
# Start simulation
sim.iteration_kill_switch = 200
sim.cost_kill_switch = 10000
sim.run()
def run_ei(run_index, SAMPLE_DOMAIN=1000):
FOLDER = "RNS%d" % SAMPLE_DOMAIN
sffx = "ei"
# Generate the main object
sim = Optimizer()
# Assign simulation properties
#if use_MAP:
# sim.hyperparameter_objective = MAP
#else:
sim.hyperparameter_objective = MLE
###################################################################################################
# File names
sim.fname_out = None
sim.fname_historical = None
sim.logger_fname = "%s/%d_%s.log" % (FOLDER, run_index, sffx)
if os.path.exists(sim.logger_fname):
os.system("rm %s" % sim.logger_fname)
os.system("touch %s" % sim.logger_fname)
sim.obj_vs_cost_fname = None
sim.mu_fname = None
sim.sig_fname = None
sim.combos_fname = None
sim.hp_fname = None
sim.acquisition_fname = None
sim.save_extra_files = True
# Information sources, in order from expensive to cheap
rosenbrock = lambda x1, x2: (1.0 - x1)**2 + 100.0 * (x2 - x1**2)**2 - 456.3
sim.IS = [lambda x1, x2: -1.0 * rosenbrock(x1, x2)]
sim.costs = [1000.0]
########################################
sim.numerical = True
sim.historical_nsample = 5
sim.domain = [(-2.0, 2.0), (-2.0, 2.0)]
sim.sample_n_from_domain = SAMPLE_DOMAIN
########################################
sim.n_start = 10 # The number of starting MLE samples
sim.reopt = 20
sim.ramp_opt = None
sim.parallel = False
# Parameters for debugging and overwritting
sim.debug = False
sim.verbose = True
sim.overwrite = True # If True, warning, else Error
# Functional forms of our mean and covariance
sim.mean = lambda X, Y, theta: np.array([-456.3 for _ in Y])
def cov(X0, Y, theta):
return squared(np.array(X0)[:, 1:], [theta.l1, theta.l2], theta.sig_1)
sim.cov = cov
sim.theta.bounds = {}
sim.theta.sig_1, sim.theta.bounds['sig_1'] = None, (1E-2,
lambda _, Y: np.var(Y))
sim.theta.l1, sim.theta.bounds['l1'] = None, (1E-1, 1)
sim.theta.l2, sim.theta.bounds['l2'] = None, (1E-1, 1)
sim.theta.rho = {
str(sorted([i, j])): 1.0
for i in range(len(sim.IS)) for j in range(i, len(sim.IS))
}
for k in sim.theta.rho.keys():
sim.theta.bounds['rho %s' % k] = (0.1, 1.0)
a, b = eval(k)
if a != b:
sim.theta.bounds['rho %s' % k] = (0.01, 1.0 - 1E-6)
sim.theta.set_hp_names()
sim.update_hp_only_with_IS0 = False
sim.update_hp_only_with_overlapped = False
sim.theta.normalize_L = False
sim.theta.normalize_Ks = False
###################################################################################################
# Start simulation
sim.iteration_kill_switch = 200
sim.cost_kill_switch = 10000
sim.run()
def run_misokg(run_index, sffx="misokg", SAMPLE_DOMAIN=1000):
FOLDER = "RNS%d" % SAMPLE_DOMAIN
scaled = False
dpc = False
invert_dpc = False
scaled = False
use_I = False
use_J = False
use_miso = False
if sffx == "misokg":
use_miso = True
elif sffx == "bdpc":
dpc = True
elif sffx == "bidpc":
dpc = True
invert_dpc = True
elif sffx == "bvl":
pass
elif sffx == "bsvl":
scaled = True
elif sffx == "bI":
use_I = True
elif sffx == "bu":
use_J = True
else:
raise Exception("This sffx (%s) is not accounted for." % sffx)
# Generate the main object
sim = Optimizer()
# Assign simulation properties
#if use_MAP:
# sim.hyperparameter_objective = MAP
#else:
sim.hyperparameter_objective = MLE
###################################################################################################
# File names
sim.fname_out = None
sim.fname_historical = None
sim.logger_fname = "%s/%d_%s.log" % (FOLDER, run_index, sffx)
if os.path.exists(sim.logger_fname):
os.system("rm %s" % sim.logger_fname)
os.system("touch %s" % sim.logger_fname)
sim.obj_vs_cost_fname = None
sim.mu_fname = None
sim.sig_fname = None
sim.combos_fname = None
sim.hp_fname = None
sim.acquisition_fname = None
sim.save_extra_files = True
# Information sources, in order from expensive to cheap
rosenbrock = lambda x1, x2: (1.0 - x1)**2 + 100.0 * (x2 - x1**2)**2 - 456.3
sim.IS = [
lambda x1, x2: -1.0 * rosenbrock(x1, x2), lambda x1, x2: -1.0 *
(rosenbrock(x1, x2) + 0.1 * np.sin(10.0 * x1 + 5.0 * x2))
]
#sim.IS = [
# lambda x1, x2: (1.0 - x1)**2 + 100.0 * (x2 - x1**2)**2 - 456.3 + np.random.normal()
# lambda x1, x2: (1.0 - x1)**2 + 100.0 * (x2 - x1**2)**2 - 456.3 + 2.0 * np.sin(10.0 * x1 + 5.0 * x2)
#]
sim.costs = [1000.0, 1.0]
########################################
sim.numerical = True
sim.historical_nsample = 5
sim.domain = [(-2.0, 2.0), (-2.0, 2.0)]
sim.sample_n_from_domain = SAMPLE_DOMAIN
########################################
sim.n_start = 10 # The number of starting MLE samples
sim.reopt = 20
sim.ramp_opt = None
sim.parallel = False
# Parameters for debugging and overwritting
sim.debug = False
sim.verbose = True
sim.overwrite = True # If True, warning, else Error
sim.acquisition = getNextSample_misokg
# Functional forms of our mean and covariance
sim.mean = lambda X, Y, theta: np.array([-456.3 for _ in Y])
def cov_miso(X0, Y, theta):
Kx = squared(np.array(X0)[:, 1:], [theta.l1, theta.l2], theta.sig_1)
Kx_l = squared(np.array(X0)[:, 1:], [theta.l3, theta.l4], theta.sig_2)
return np.block([[Kx, Kx], [Kx, Kx + Kx_l]])
def cov_bonilla(X0, Y, theta):
Kx = squared(np.array(X0)[:, 1:], [theta.l1, theta.l2], theta.sig_1)
Kx = Kx + 1E-6 * np.eye(Kx.shape[0])
if use_J:
Ks = np.ones((theta.n_IS, theta.n_IS)) * (1.0 - 1E-6) + np.eye(
theta.n_IS) * 1E-6
elif use_I:
Ks = np.eye(theta.n_IS)
elif dpc and invert_dpc:
Ks = np.array([
np.array([
1.0 if i != j else theta.rho["[0, %d]" % i]**(-2.0)
for j in range(theta.n_IS)
]) for i in range(theta.n_IS)
])
elif dpc:
Ks = np.array([
np.array([
theta.rho[str(sorted([i, j]))] for j in range(theta.n_IS)
]) for i in range(theta.n_IS)
])
else:
L = np.array([
np.array([
theta.rho[str(sorted([i, j]))] if i >= j else 0.0
for j in range(theta.n_IS)
]) for i in range(theta.n_IS)
])
# Force it to be positive semi-definite
Ks = L.dot(L.T)
if theta.n_IS == 2:
e = np.diag(np.array([theta.e1, theta.e2]))
elif theta.n_IS == 3:
e = np.diag(np.array([theta.e1, theta.e2, theta.e3]))
else:
raise Exception("HOW?")
Ks = e.dot(Ks.dot(e))
return np.kron(Ks, Kx)
sim.theta.bounds = {}
sim.theta.sig_1, sim.theta.bounds['sig_1'] = None, (1E-2,
lambda _, Y: np.var(Y))
sim.theta.l1, sim.theta.bounds['l1'] = None, (1E-1, 1)
sim.theta.l2, sim.theta.bounds['l2'] = None, (1E-1, 1)
if use_miso:
sim.cov = cov_miso
sim.theta.sig_2, sim.theta.bounds['sig_2'] = None, (
1E-2, lambda _, Y: np.var(Y))
sim.theta.l3, sim.theta.bounds['l3'] = None, (1E-1, 1)
sim.theta.l4, sim.theta.bounds['l4'] = None, (1E-1, 1)
sim.theta.rho = {
str(sorted([i, j])): 1.0
for i in range(len(sim.IS)) for j in range(i, len(sim.IS))
}
else:
sim.cov = cov_bonilla
if scaled:
sim.theta.e1, sim.theta.bounds['e1'] = None, (1E-1, 1.0)
sim.theta.e2, sim.theta.bounds['e2'] = None, (1E-1, 1.0)
else:
sim.theta.e1, sim.theta.bounds['e1'] = 1.0, (1E-1, 1.0)
sim.theta.e2, sim.theta.bounds['e2'] = 1.0, (1E-1, 1.0)
sim.theta.rho = {"[0, 0]": None, "[0, 1]": None, "[1, 1]": None}
if dpc or use_I or use_J:
sim.theta.rho = {
str(sorted([i, j])): 1.0
for i in range(len(sim.IS)) for j in range(i, len(sim.IS))
}
sim.dynamic_pc = dpc
for k in sim.theta.rho.keys():
sim.theta.bounds['rho %s' % k] = (0.1, 1.0)
a, b = eval(k)
if a != b:
sim.theta.bounds['rho %s' % k] = (0.01, 1.0 - 1E-6)
sim.theta.set_hp_names()
sim.update_hp_only_with_IS0 = False
sim.update_hp_only_with_overlapped = False
sim.theta.normalize_L = False
sim.theta.normalize_Ks = False
###################################################################################################
# Start simulation
sim.iteration_kill_switch = 200
sim.cost_kill_switch = 10000
#sim.cost_kill_switch = sim.iteration_kill_switch * sim.costs[0]
sim.run()