def test_isotropic_zero_mean_equals_log_gaussian_pdf():
D = 2
x = np.random.randn(D)
g = IsotropicZeroMeanGaussian(sigma=np.sqrt(2))
log_pdf = log_gaussian_pdf(x,
mu=np.zeros(D),
Sigma=np.eye(D) * 2,
is_cholesky=False,
compute_grad=False)
assert_close(log_pdf, g.log_pdf(x))
def habc_generator(num_warmup, thin_step):
D = 10
step_size_min = 0.01
step_size_max = .1
num_steps_min = 50
num_steps_max = 50
sigma_p = 1.
momentum_seed = np.random.randint(time.time())
momentum = IsotropicZeroMeanGaussian(sigma=sigma_p, D=D)
abc_target = ABCSkewNormalPosterior(theta_true=np.ones(D) * 10)
start = abc_target.theta_true
job = HABCJob(abc_target, momentum, num_iterations, start, num_steps_min,
num_steps_max, step_size_min, step_size_max, momentum_seed,
statistics, num_warmup, thin_step)
job.walltime = 60 * 60
# store results in home dir straight away
d = os.sep.join(os.path.abspath(__file__).split(os.sep)[:-1]) + os.sep
job.aggregator = MCMCJobResultAggregatorStoreHome(d)
return job
def kmc_generator(num_warmup, thin_step):
D = 10
step_size_min = 0.01
step_size_max = .1
num_steps_min = 50
num_steps_max = 50
sigma_p = 1.
momentum_seed = np.random.randint(time.time())
momentum = IsotropicZeroMeanGaussian(sigma=sigma_p, D=D)
abc_target = ABCSkewNormalPosterior(theta_true=np.ones(D) * 10)
start = abc_target.theta_true
Z = np.load("../ground_truth/benchmark_samples.arr")[:1000]
learn_parameters = False
force_relearn_parameters = False
lmbda = 1.
sigma = 2**4
if False and True:
sigma = select_sigma_grid(Z)
select_sigma_lambda_cma(Z,
num_folds=5,
num_repetitions=1,
sigma0=0.31,
lmbda0=1.)
exit()
logger.info("Using sigma=%.6f" % sigma)
job = KMCLiteJob(Z, sigma, lmbda, abc_target, momentum, num_iterations,
start, num_steps_min, num_steps_max, step_size_min,
step_size_max, momentum_seed, learn_parameters,
force_relearn_parameters, statistics, num_warmup,
thin_step)
# marginal sampler
job.recompute_log_pdf = True
job.walltime = 60 * 60
# store results in home dir straight away
d = os.sep.join(os.path.abspath(__file__).split(os.sep)[:-1]) + os.sep
job.aggregator = MCMCJobResultAggregatorStoreHome(d)
return job
def kmc_generator(num_warmup, thin_step):
D = 9
start = np.random.randn(D) * 0
step_size_min = 0.02
step_size_max = 0.2
num_steps_min = 50
num_steps_max = 100
sigma_p = 1.
momentum_seed = np.random.randint(time.time())
momentum = IsotropicZeroMeanGaussian(sigma=sigma_p, D=D)
# estimator parameters (comments are x-validation scores on benchmark samples for lmbda = 0.000001 and m=1000)
# sigma = 0.5 # -10.5821369355, -10.621941424
sigma = 0.6 # -11.1362549964, -11.2676576859, -11.0283705786
# sigma = 0.65 # -11.2538670706, -11.247122781, -10.9686599646
# sigma = 0.7 # -11.052076379, -11.1454946033, -10.7527806709
# sigma = 0.8 # -10.2950464716, -10.6814600267
lmbda = 0.000001
target = GlassPosterior()
m = 1000
Z = np.load("benchmark_samples.arr")[:m]
learn_parameters = False
force_relearn_parameters = False
job = KMCRandomFeatsJob(Z, m, sigma, lmbda, target, momentum,
num_iterations, start, num_steps_min,
num_steps_max, step_size_min, step_size_max,
momentum_seed, learn_parameters,
force_relearn_parameters, statistics, num_warmup,
thin_step)
job.walltime = 60 * 60
# store results in home dir straight away
d = os.sep.join(os.path.abspath(__file__).split(os.sep)[:-1]) + os.sep
job.aggregator = MCMCJobResultAggregatorStoreHome(d)
return job
def kmc_generator(N, D, target, num_warmup, thin_step, momentum_seed):
if D == 2:
step_size_min = 0.8
step_size_max = 1.5
elif D == 8:
step_size_min = 0.6
step_size_max = 1.3
start = get_start(D)
momentum = IsotropicZeroMeanGaussian(sigma=sigma_p, D=D)
# estimator parameters
sigma = 0.46
lmbda = 0.000001
learn_parameters = True if N < 500 else False
force_relearn_parameters = True if N < 500 else False
# oracle samples
Z = sample_banana(N, D, bananicity, V)
job = KMCRandomFeatsJob(Z,
N,
sigma,
lmbda,
target,
momentum,
num_iterations,
start,
num_steps_min,
num_steps_max,
step_size_min,
step_size_max,
momentum_seed,
learn_parameters=learn_parameters,
force_relearn_parameters=force_relearn_parameters,
statistics=statistics,
num_warmup=num_warmup,
thin_step=thin_step)
job.plot = False
return job
def kmc_generator(num_warmup, thin_step):
D = 9
start = np.random.randn(D) * 0
step_size_min = 0.01
step_size_max = 0.1
num_steps_min = 1
num_steps_max = 10
sigma_p = 1.
momentum_seed = np.random.randint(time.time())
momentum = IsotropicZeroMeanGaussian(sigma=sigma_p, D=D)
target = GlassPosterior()
Z = np.load("../ground_truth/benchmark_samples.arr")[:1000]
learn_parameters = False
force_relearn_parameters = False
lmbda = 1.
if False:
sigma = select_sigma_grid(Z, lmbda=lmbda, plot_surface=True)
sigma = 2**4
job = KMCLiteJob(Z, sigma, lmbda, target, momentum, num_iterations, start,
num_steps_min, num_steps_max, step_size_min,
step_size_max, momentum_seed, learn_parameters,
force_relearn_parameters, statistics, num_warmup,
thin_step)
job.walltime = 60 * 60
# store results in home dir straight away
d = os.sep.join(os.path.abspath(__file__).split(os.sep)[:-1]) + os.sep
job.aggregator = MCMCJobResultAggregatorStoreHome(d)
return job
def hmc_generator(D, target, num_warmup, thin_step, momentum_seed):
# determined by pilot runs
if D == 2:
step_size_min = 0.8
step_size_max = 1.5
elif D == 8:
step_size_min = 0.6
step_size_max = 1.3
start = get_start(D)
momentum = IsotropicZeroMeanGaussian(sigma=sigma_p, D=D)
return HMCJob(target,
momentum,
num_iterations,
start,
num_steps_min,
num_steps_max,
step_size_min,
step_size_max,
momentum_seed,
statistics=statistics,
num_warmup=num_warmup,
thin_step=thin_step)
def kmc_generator(N, D, target, num_warmup, thin_step, momentum_seed):
if D == 2:
step_size_min = 0.8
step_size_max = 1.5
elif D == 8:
step_size_min = 0.6
step_size_max = 1.3
start = get_start(D)
momentum = IsotropicZeroMeanGaussian(sigma=sigma_p, D=D)
learn_parameters = False
force_relearn_parameters = False
# oracle samples
Z = sample_banana(N, D, bananicity, V=100)
# estimator parameters
lmbda = 1.
# sigma = select_sigma_grid(Z, lmbda=lmbda, plot_surface=True)
sigma = 50.80
job = KMCLiteJob(Z,
sigma,
lmbda,
target,
momentum,
num_iterations,
start,
num_steps_min,
num_steps_max,
step_size_min,
step_size_max,
momentum_seed,
learn_parameters=learn_parameters,
force_relearn_parameters=force_relearn_parameters,
statistics=statistics,
num_warmup=num_warmup,
thin_step=thin_step)
# job.set_up()
#
#
# res = 50
# Xs = np.linspace(-15, 15,res)
# Ys = np.linspace(-10, 5,res)
#
# # evaluate density and estimate
# D1=0
# D2=1
# def dummy_grad(X_2d):
# theta = start.copy()
# # theta = np.mean(self.Z, 0)
# theta[D1]=X_2d[0]
# theta[D2]=X_2d[1]
# return job.target.grad(theta)
#
# def dummy(X_2d):
# theta = start.copy()
# # theta = np.mean(self.Z, 0)
# theta[D1]=X_2d[0]
# theta[D2]=X_2d[1]
# return job.target.log_pdf(theta)
#
# import matplotlib.pyplot as plt
# plt.figure()
# G = evaluate_density_grid(Xs, Ys, dummy)
# plot_array(Xs, Ys, G)
# plt.plot(Z[:,D1], Z[:,D2], '.')
# plt.plot(start[D1], start[D2], 'b*', markersize=15)
#
# plt.figure()
# G_norm, U, V, X, Y = evaluate_gradient_grid(Xs, Ys, dummy_grad)
# plot_array(Xs, Ys, G_norm)
# plt.plot(Z[:,D1], Z[:,D2], '.')
# plt.quiver(X, Y, U, V, color='m')
# plt.show()
job.plot = False
return job