from kameleon_rks.experiments.tools import store_samples
result_fname = os.path.splitext(os.path.basename(__file__))[0] + ".txt"
if __name__ == '__main__':
import time
from kameleon_rks.examples.plotting import visualise_pairwise_marginals
from kameleon_rks.proposals.Metropolis import AdaptiveIndependentMetropolis
from kameleon_rks.samplers.mini_pmc import mini_pmc
from kameleon_rks.tools.log import Log
import numpy as np
from smc2.sv_models import SVoneSP500Model
logger = Log.get_logger()
def get_AdaptiveIndependentMetropolis_instance(D, target_log_pdf):
gamma2 = 0.1
# # run 1 (initial parameters from Ingmar's pilot runs)
# proposal_mu = np.array([ -0.17973298, 0.11796741, 0.86733172, 0.52834129, -3.32354247])
# proposal_var = np.array([ 0.01932092, 0.02373668, 0.02096583, 0.1566503 , 0.46316933])
# proposal_L_C = np.diag(np.sqrt(proposal_var))
# proposal_L_C *= 2
# # resuts:
# # mean: array([ 0.08820961, -0.03436691, 0.7178945, 0.54124475, -3.77499049])
# # var: array([ 0.01514925, 0.01407534, 0.056966, 0.37502436, 0.5887545])
# # np.mean(var): 0.201
# # np.linalg.norm(mean): 3.882
import time
from kameleon_rks.tools.log import Log
import numpy as np
logger = Log.get_logger()
def mini_mcmc(transition_kernel,
start,
num_iter,
D,
recompute_log_pdf=False,
time_budget=None):
# MCMC results
samples = np.zeros((num_iter, D)) + np.nan
proposals = np.zeros((num_iter, D)) + np.nan
accepted = np.zeros(num_iter) + np.nan
acc_prob = np.zeros(num_iter) + np.nan
log_pdf = np.zeros(num_iter) + np.nan
step_sizes = np.zeros(num_iter) + np.nan
# timings for output and time limit
times = np.zeros(num_iter)
last_time_printed = time.time()
# for adaptive transition kernels
avg_accept = 0.
# init MCMC (first iteration)
current = start
if False:
plt.scatter(samples.T[0],
samples.T[1],
c='r',
marker='*',
zorder=4,
s=5)
# fig.suptitle("%s - importance resampled" % (sampler_is.__class__.__name__,))
plt.show()
plt.scatter(unadj_samp.T[0],
unadj_samp.T[1],
c=logw - logsumexp(logw),
cmap=plt.get_cmap('Blues'),
alpha=0.5,
zorder=2) #)visualize_scatter_2d()
# plt.suptitle("%s - unadjusted Langevin" % (sampler_is.__class__.__name__,))
# plt.scatter(mcmc_samps[0].T[0], mcmc_samps[0].T[1], c='b',marker='*')
plt.show()
Log.get_logger().info('====' + str(sampler_mh.step_size) + ' ' +
str(mcmc_samps[2].mean()) + '====')
#the following two should be 0 ideally
#mmd_w = pk.estimateMMD(Z,samples)
#mmd_unw = pk.estimateMMD(Z,unadj_samp)
Log.get_logger().info(
'estimated weighted mse moments %.2f, res per gen %.2f unadj %.2f, mcmc %.2f\n weight/unw. (good if <1):%.2f\n is/mcmc %2f'
% (mom_w, mom_gen, mom_unadj, mom_mcmc, mom_w / mom_unadj,
mom_w / mom_mcmc))
# plt.show()
D = mdl.dim
pop_size=50
target_log_pdf = mdl.get_logpdf_closure()
samplers = [
# get_StaticMetropolis_instance(D, target_log_pdf),
# get_AdaptiveMetropolis_instance(D, target_log_pdf),
# get_OracleKameleon_instance(D, target_log_pdf),
# get_Kameleon_instance(D, target_log_pdf),
# get_StaticLangevin_instance(D, target_log_pdf, target_grad),
# get_AM_5(D, target_log_pdf),
get_AM_1(D, target_log_pdf),
# get_AM_2(D, target_log_pdf),
# get_AM_05(D, target_log_pdf),
]
for sampler in samplers:
print(sampler.__class__)
start = mdl.rvs(pop_size)
num_iter = 5000
samples, log_target_densities, times = mini_pmc(sampler, start, num_iter, pop_size)
mom_samp = np.array([(samples**i).mean(0) for i in range(1,4)])
visualize_scatter_2d(samples[:,:2])
Log.get_logger().info('===='+str(sampler.step_size)+'====')
plt.show()
target_grad = lambda x: log_banana_pdf(x, bananicity, V, compute_grad=True)
samplers = [
# get_StaticMetropolis_instance(D, target_log_pdf),
# get_AdaptiveMetropolis_instance(D, target_log_pdf),
# get_OracleKameleon_instance(D, target_log_pdf),
# get_Kameleon_instance(D, target_log_pdf),
# get_StaticLangevin_instance(D, target_log_pdf, target_grad),
get_AM_5(D, target_log_pdf),
get_AM_1(D, target_log_pdf),
get_AM_2(D, target_log_pdf),
get_AM_05(D, target_log_pdf),
]
for sampler in samplers:
start = np.zeros(D)
num_iter = 1000
samples, log_target_densities, times = mini_pmc(sampler, start, num_iter, pop_size)
mom_samp = np.array([(samples**i).mean(0) for i in range(1,4)])
visualize_scatter_2d(samples)
Log.get_logger().info('===='+str(sampler.step_size)+'====')
#the following two should be 0 ideally
Log.get_logger().info(np.mean((esj(samples,pop_size)-1)**2))
Log.get_logger().info(np.mean(((mom_samp / moments)-1)**2))
plt.suptitle("%s" % \
(sampler.__class__.__name__,))
plt.show()
D = mdl.dim
pop_size = 50
target_log_pdf = mdl.get_logpdf_closure()
samplers = [
# get_StaticMetropolis_instance(D, target_log_pdf),
# get_AdaptiveMetropolis_instance(D, target_log_pdf),
# get_OracleKameleon_instance(D, target_log_pdf),
# get_Kameleon_instance(D, target_log_pdf),
# get_StaticLangevin_instance(D, target_log_pdf, target_grad),
# get_AM_5(D, target_log_pdf),
get_AM_1(D, target_log_pdf),
# get_AM_2(D, target_log_pdf),
# get_AM_05(D, target_log_pdf),
]
for sampler in samplers:
print(sampler.__class__)
start = mdl.rvs(pop_size)
num_iter = 5000
samples, log_target_densities, times = mini_pmc(
sampler, start, num_iter, pop_size)
mom_samp = np.array([(samples**i).mean(0) for i in range(1, 4)])
visualize_scatter_2d(samples[:, :2])
Log.get_logger().info('====' + str(sampler.step_size) + '====')
plt.show()
samples, log_target_densities, unadj_samp, unadj_log_target, logw, unw_samptimes = mini_rb_pmc(sampler_is, start, num_iter, pop_size, D, time_budget=100000)
mom_gen = np.mean((np.array([(samples**i).mean(0) for i in range(1, max_moment)]) - moments)**2)
mcmc_samps = mini_mcmc(sampler_mh, start, num_iter, D)
#the weights we get back are not Rao-Blackwellized, which is what we do now.
#beware: this only works if the proposal is not adapted during sampling!!
#logw = logsumexp(np.array([sampler_is.proposal_log_pdf(i, unadj_samp) for i in unadj_samp]), 0)
res_idx = system_res(range(len(logw)), logw, resampled_size=10*len(logw))
samples = unadj_samp[res_idx]
mom_unadj = np.mean((np.array([(unadj_samp**i).mean(0) for i in range(1, max_moment)]) - moments)**2)
mom_w = np.mean((np.array([(unadj_samp**i * exp(logw - logsumexp(logw))[:,np.newaxis]).sum(0) for i in range(1, max_moment)]) -moments)**2)
mom_mcmc = np.mean((np.array([(mcmc_samps[0]**i).mean(0) for i in range(1, max_moment)]) - moments)**2)
if False:
plt.scatter(samples.T[0], samples.T[1], c='r', marker='*', zorder=4, s=5)
# fig.suptitle("%s - importance resampled" % (sampler_is.__class__.__name__,))
plt.show()
plt.scatter(unadj_samp.T[0], unadj_samp.T[1], c = logw - logsumexp(logw), cmap = plt.get_cmap('Blues'), alpha=0.5, zorder=2) #)visualize_scatter_2d()
# plt.suptitle("%s - unadjusted Langevin" % (sampler_is.__class__.__name__,))
# plt.scatter(mcmc_samps[0].T[0], mcmc_samps[0].T[1], c='b',marker='*')
plt.show()
Log.get_logger().info('===='+str(sampler_mh.step_size)+' '+str(mcmc_samps[2].mean())+'====')
#the following two should be 0 ideally
#mmd_w = pk.estimateMMD(Z,samples)
#mmd_unw = pk.estimateMMD(Z,unadj_samp)
Log.get_logger().info('estimated weighted mse moments %.2f, res per gen %.2f unadj %.2f, mcmc %.2f\n weight/unw. (good if <1):%.2f\n is/mcmc %2f' % (mom_w, mom_gen, mom_unadj,mom_mcmc, mom_w/mom_unadj, mom_w/mom_mcmc))
# plt.show()