def fit_beta_by_pt(nsteps):
"""Fit a beta distribution by parallel tempering."""
num_dimensions = 1
# Create the dummy model
b = BetaFit(0.5, 0.5)
# Create the options
opts = MCMCOpts()
opts.model = b
opts.estimate_params = b.parameters
opts.initial_values = [10 ** 0.5]
opts.nsteps = nsteps
opts.anneal_length = 0
opts.T_init = 1
opts.use_hessian = False
opts.seed = 1
opts.norm_step_size = 0.5
opts.likelihood_fn = b.likelihood
opts.step_fn = step
# Create the MCMC object
num_temps = 8
pt = PT_MCMC(opts, num_temps, 10)
pt.estimate()
plt.ion()
for chain in pt.chains:
fig = plt.figure()
chain.prune(nsteps/10, 1)
(heights, points, lines) = plt.hist(chain.positions, bins=100,
normed=True)
plt.plot(points, beta.pdf(points, b.a, b.b), 'r')
plt.ylim((0,10))
plt.xlim((0, 1))
return pt
def fit_beta(nsteps):
"""Fit a beta distribution by MCMC."""
num_dimensions = 1
# Create the dummy model
b = BetaFit(0.5, 0.5)
# Create the options
opts = MCMCOpts()
opts.model = b
opts.estimate_params = b.parameters
opts.initial_values = [1.001]
opts.nsteps = nsteps
opts.anneal_length = nsteps/10
opts.T_init = 100
opts.use_hessian = False
opts.seed = 1
opts.norm_step_size = 0.01
opts.likelihood_fn = b.likelihood
opts.step_fn = step
# Create the MCMC object
mcmc = MCMC(opts)
mcmc.initialize()
mcmc.estimate()
mcmc.prune(nsteps/10, 1)
plt.ion()
for i in range(mcmc.num_estimate):
plt.figure()
(heights, points, lines) = plt.hist(mcmc.positions[:,i], bins=100,
normed=True)
plt.plot(points, beta.pdf(points, b.a, b.b), 'r')
return mcmc
def fit_gaussian(nsteps):
num_dimensions = 1
# Create the dummy model
means = 10 ** np.random.rand(num_dimensions)
variances = np.random.rand(num_dimensions)
g = GaussianFit(means, variances)
# Create the options
opts = MCMCOpts()
opts.model = g
opts.estimate_params = g.parameters
opts.initial_values = [1]
opts.nsteps = nsteps
opts.anneal_length = nsteps/10
opts.T_init = 1
opts.use_hessian = False
opts.seed = 1
opts.norm_step_size = 1
opts.likelihood_fn = g.likelihood
opts.step_fn = step
# Create the MCMC object
mcmc = MCMC(opts)
mcmc.initialize()
mcmc.estimate()
mcmc.prune(nsteps/10, 1)
plt.ion()
for i in range(mcmc.num_estimate):
mean_i = np.mean(mcmc.positions[:,i])
var_i = np.var(mcmc.positions[:, i])
print "True mean: %f" % means[i]
print "Sampled mean: %f" % mean_i
print "True variance: %f" % variances[i]
print "Sampled variance: %f" % var_i
plt.figure()
(heights, points, lines) = plt.hist(mcmc.positions[:,i], bins=50,
normed=True)
plt.plot(points, norm.pdf(points, loc=means[i],
scale=np.sqrt(variances[i])), 'r')
mean_err = np.zeros(len(mcmc.positions))
var_err = np.zeros(len(mcmc.positions))
for i in range(len(mcmc.positions)):
mean_err[i] = (means[0] - np.mean(mcmc.positions[:i+1,0]))
var_err[i] = (variances[0] - np.var(mcmc.positions[:i+1,0]))
plt.figure()
plt.plot(mean_err)
plt.plot(var_err)
return mcmc
def fit_twod_gaussians_by_pt(nsteps):
means_x = [ 0.1, 0.5, 0.9,
0.1, 0.5, 0.9,
0.1, 0.5, 0.9]
means_y = [0.1, 0.1, 0.1,
0.5, 0.5, 0.5,
0.9, 0.9, 0.9]
sd = 0.01
tdg = TwoDGaussianFit(means_x, means_y, sd ** 2)
# Create the options
opts = MCMCOpts()
opts.model = tdg
opts.estimate_params = tdg.parameters
opts.initial_values = [1.001, 1.001]
opts.nsteps = nsteps
opts.anneal_length = 0 # necessary so cooling does not occur
opts.T_init = 1
opts.use_hessian = False
opts.seed = 1
opts.norm_step_size = 0.1
opts.likelihood_fn = tdg.likelihood
opts.step_fn = step
# Create the PT object
num_temps = 8
pt = PT_MCMC(opts, num_temps, 100)
pt.estimate()
plt.ion()
for chain in pt.chains:
fig = plt.figure()
chain.prune(nsteps/10, 1)
plt.scatter(chain.positions[:,0], chain.positions[:,1])
ax = fig.gca()
for x, y in zip(means_x, means_y):
circ = plt.Circle((x, y), radius=2*sd, color='r', fill=False)
ax.add_patch(circ)
plt.xlim((-0.5, 1.5))
plt.ylim((-0.5, 1.5))
plt.title('Temp = %.2f' % chain.options.T_init)
plt.show()
return pt
def fit_twod_gaussians(nsteps):
means_x = [ 0.1, 0.5, 0.9,
0.1, 0.5, 0.9,
0.1, 0.5, 0.9]
means_y = [0.1, 0.1, 0.1,
0.5, 0.5, 0.5,
0.9, 0.9, 0.9]
sd = 0.01
tdg = TwoDGaussianFit(means_x, means_y, sd ** 2)
# Create the options
opts = MCMCOpts()
opts.model = tdg
opts.estimate_params = tdg.parameters
opts.initial_values = [1.001, 1.001]
opts.nsteps = nsteps
opts.anneal_length = nsteps/10
opts.T_init = 1
opts.use_hessian = False
opts.seed = 1
opts.norm_step_size = 0.1
opts.likelihood_fn = tdg.likelihood
opts.step_fn = step
mcmc = MCMC(opts)
mcmc.initialize()
mcmc.estimate()
plt.ion()
fig = plt.figure()
mcmc.prune(0, 20)
plt.scatter(mcmc.positions[:,0], mcmc.positions[:,1])
ax = fig.gca()
for x, y in zip(means_x, means_y):
circ = plt.Circle((x, y), radius=2*sd, color='r', fill=False)
ax.add_patch(circ)
plt.xlim((-0.5, 1.5))
plt.ylim((-0.5, 1.5))
plt.show()
return mcmc
# Create temperature array based on number of workers (excluding master)
temps = np.logspace(np.log10(min_temp), np.log10(max_temp), num_chains-1)
rate_step_sizes = np.logspace(np.log10(5e-4), np.log10(5e-2),
num_chains-1)
scaling_step_sizes = np.logspace(np.log10(1e-4), np.log10(1e-2),
num_chains-1)
# Initialize the MCMC arguments
b = args['builder']
opts = MCMCOpts()
opts.model = b.model
opts.tspan = args['time']
opts.estimate_params = b.estimate_params
opts.initial_values = b.random_initial_values()
opts.nsteps = args['nsteps']
#opts.norm_step_size = np.array([0.01] + \
# ([0.05] * (len(opts.estimate_params)-1)))
# FIXME not correct ordering
#opts.norm_step_size = np.array([scaling_step_sizes[rank-1]] + \
# ([rate_step_sizes[rank-1]] *
# (len(opts.estimate_params)-1)))
#print "Step size: %g" % (opts.norm_step_size[1])
opts.norm_step_size = 0.05
opts.sigma_step = 0 # Don't adjust step size
#opts.sigma_max = 50
#opts.sigma_min = 0.01
#opts.accept_rate_target = 0.23
#opts.accept_window = 100
0.1, 0.5, 0.9]
means_y = [0.1, 0.1, 0.1,
0.5, 0.5, 0.5,
0.9, 0.9, 0.9]
sd = 0.01
tdg = TwoDGaussianFit(means_x, means_y, sd ** 2)
# Create temperature array based on number of workers (excluding master)
temps = np.logspace(np.log10(min_temp), np.log10(max_temp), num_chains-1)
# Create the options
opts = MCMCOpts()
opts.model = tdg
opts.estimate_params = tdg.parameters
opts.initial_values = [1.001, 1.001]
opts.nsteps = nsteps
opts.anneal_length = 0 # necessary so cooling does not occur
opts.T_init = temps[rank - 1] # Use the temperature for this worker
opts.use_hessian = False
opts.seed = 1
opts.norm_step_size = 0.1
opts.likelihood_fn = tdg.likelihood
mcmc = MCMC(opts)
mcmc.initialize()
# The master coordinates when swaps occur ---------
if rank == 0:
pt = PT_MPI_Master(comm, rank, opts, swap_period, num_chains)
pt.run()
# Everyone else runs MCMC steps and swaps when told -----------
# Set up the 2-D gaussian model
means_x = [0.1, 0.5, 0.9, 0.1, 0.5, 0.9, 0.1, 0.5, 0.9]
means_y = [0.1, 0.1, 0.1, 0.5, 0.5, 0.5, 0.9, 0.9, 0.9]
sd = 0.01
tdg = TwoDGaussianFit(means_x, means_y, sd**2)
# Create temperature array based on number of workers (excluding master)
temps = np.logspace(np.log10(min_temp), np.log10(max_temp), num_chains - 1)
# Create the options
opts = MCMCOpts()
opts.model = tdg
opts.estimate_params = tdg.parameters
opts.initial_values = [1.001, 1.001]
opts.nsteps = nsteps
opts.anneal_length = 0 # necessary so cooling does not occur
opts.T_init = temps[rank - 1] # Use the temperature for this worker
opts.use_hessian = False
opts.seed = 1
opts.norm_step_size = 0.1
opts.likelihood_fn = tdg.likelihood
mcmc = MCMC(opts)
mcmc.initialize()
# The master coordinates when swaps occur ---------
if rank == 0:
pt = PT_MPI_Master(comm, rank, opts, swap_period, num_chains)
pt.run()
# Everyone else runs MCMC steps and swaps when told -----------