model = SpringMassModel(state0, measure_t_grid)
# Load data
noise_stddev = 0.5
displacement_data = np.genfromtxt('noisy_data.txt')
# Define prior distributions
param_priors = {'K': ['Uniform', 0.0, 10.0], 'g': ['Uniform', 0.0, 10.0]}
# SMC sampling
num_particles = 500
num_time_steps = 20
num_mcmc_steps = 1
# Autosaver
smc = SMCSampler(displacement_data, model, param_priors)
step_list = smc.sample(num_particles,
num_time_steps,
num_mcmc_steps,
noise_stddev,
ess_threshold=num_particles * 0.5,
autosave_file='autosaver.hdf5')
try:
# Calculate means
means = [step.get_mean() for step in step_list]
for key, value in means[-1].iteritems():
print '%s mean = %s' % (key, value)
# Plot
if __name__ == '__main__':
# instance model / set up ground truth / add noise
a = 2
b = 3.5
x = np.arange(50)
m = model(x)
std_dev = 0.6
y_noisy = np.genfromtxt('noisy_data.txt')
param_priors = {'a': ['Uniform', -5.0, 5.0], 'b': ['Uniform', -5.0, 5.0]}
# run smc
num_particles = 1000
num_time_steps = 20
num_mcmc_steps = 5
smc = SMCSampler(y_noisy, m, param_priors)
particle_chain = smc.sample(num_particles,
num_time_steps,
num_mcmc_steps,
std_dev,
ess_threshold=0.5 * num_particles,
autosave_file='smc.h5')
# try a restart
restarted_chain = smc.sample(num_particles,
num_time_steps,
num_mcmc_steps,
std_dev,
ess_threshold=0.5 * num_particles,
restart_time_step=10,
hdf5_to_load='smc.h5',
import numpy as np
from spring_mass_models import SpringMassModel
from smcpy.smc.smc_sampler import SMCSampler
# Initialize model
state0 = [0., 0.] #initial conditions
measure_t_grid = np.arange(0., 5., 0.2) #time
model = SpringMassModel(state0, measure_t_grid)
# Load data
noise_stddev = 0.94
displacement_data = np.genfromtxt('noisy_data.txt')
# Define prior distributions
param_priors = {'K': ['Uniform', 0.0, 10.0], 'g': ['Uniform', 0.0, 10.0]}
# SMC sampling
num_particles = 5000
num_time_steps = 20
num_mcmc_steps = 1
smc = SMCSampler(displacement_data, model, param_priors)
pchain = smc.sample(num_particles,
num_time_steps,
num_mcmc_steps,
noise_stddev,
ESS_threshold=num_particles * 0.5)
if smc.rank == 0:
pchain.plot_pairwise_weights(save=True)
from scipy.optimize import minimize
if __name__ == '__main__':
# instance model / set up ground truth / add noise
a = 2
b = 3.5
x = np.arange(50)
m = model(x)
std_dev = 0.6
y_noisy = np.genfromtxt('noisy_data.txt')
param_priors = {'a': ['Uniform', -5.0, 5.0], 'b': ['Uniform', -5.0, 5.0]}
# run smc
num_particles = 1000
num_time_steps = 20
num_mcmc_steps = 5
smc = SMCSampler(y_noisy, m, param_priors)
particle_chain = smc.sample(
num_particles,
num_time_steps,
num_mcmc_steps,
std_dev,
ESS_threshold=0.5 * num_particles,
#proposal_center=center, proposal_scales=scales,
autosave_file='test.h5')
try:
particle_chain.plot_pairwise_weights()
except:
pass
def test_setup_communicator():
comm, size, my_rank = SMCSampler.setup_communicator()
assert size == 1 and my_rank == 0
def test_setup_mcmc_sampler(model):
data = model.evaluate({'a': 0., 'b': 0.})
param_priors = {'a': ['Uniform', 0., 1.], 'b': ['Uniform', 0., 1.]}
mcmc = SMCSampler.setup_mcmc_sampler(data, model, param_priors)
assert isinstance(mcmc, MCMCSampler)
from scipy.optimize import minimize
if __name__ == '__main__':
# instance model / set up ground truth / add noise
a = 2
b = 3.5
x = np.arange(50)
my_model = Model(x)
std_dev = None # measurement noise std deviation will be sampled
noisy_data = np.genfromtxt('noisy_data.txt')
param_priors = {'a': ['Uniform', -5.0, 5.0], 'b': ['Uniform', -5.0, 5.0]}
# run smc
num_particles = 1000
num_time_steps = 20
num_mcmc_steps = 2
smc = SMCSampler(noisy_data, my_model, param_priors)
step_list = smc.sample(num_particles,
num_time_steps,
num_mcmc_steps,
std_dev,
ess_threshold=0.5 * num_particles,
autosave_file='test.h5')
# plot results of last step
try:
step_list[-1].plot_pairwise_weights(show=False, save=True)
except:
pass