def test_austerity(self):
np.random.seed(SEED)
X = gen_X(SAMPLE_SIZE)
def vectorized_log_lik(X,theta):
return _vector_of_log_likelihoods(theta[0],theta[1],X)
def log_density_prior(theta):
return np.log(norm.pdf(theta[0],0, SIGMA_1)) + np.log(norm.pdf(theta[1],0, SIGMA_2))
sample,_ = austerity(vectorized_log_lik,log_density_prior, X,0.01,batch_size=50,chain_size=10, thinning=1, theta_t=np.random.randn(2))
assert_almost_equal(np.array([-0.2554517, 1.3805683]),sample[-1])
def test_austerity(self):
np.random.seed(SEED)
X = gen_X(SAMPLE_SIZE)
def vectorized_log_lik(X, theta):
return _vector_of_log_likelihoods(theta[0], theta[1], X)
def log_density_prior(theta):
return np.log(norm.pdf(theta[0], 0, SIGMA_1)) + np.log(
norm.pdf(theta[1], 0, SIGMA_2))
sample, _ = austerity(vectorized_log_lik,
log_density_prior,
X,
0.01,
batch_size=50,
chain_size=10,
thinning=1,
theta_t=np.random.randn(2))
assert_almost_equal(np.array([-0.2554517, 1.3805683]), sample[-1])
from sampplers.MetropolisHastings import metropolis_hastings
import numpy as np
np.random.seed(SEED)
X = gen_X(SAMPLE_SIZE)
def vectorized_log_lik(X,theta):
return _vector_of_log_likelihoods(theta[0],theta[1],X)
def log_density_prior(theta):
return np.log(norm.pdf(theta[0],0, SIGMA_1)) + np.log(norm.pdf(theta[1],0, SIGMA_2))
sample = austerity(vectorized_log_lik,log_density_prior, X,0.01,batch_size=50,chain_size=20*1000, thinning=1, theta_t=np.random.randn(2))
print(acf(sample[:,1],nlags=50))
#
# import seaborn as sns
# sns.set(color_codes=True)
# with sns.axes_style("white"):
# pr = sns.jointplot(x=sample[:,0], y=sample[:,1], kind="kde", color="k");
#
# # pr.savefig('../../write_up/img/mcmc_sample.pdf')
#
# sns.plt.show()
def get_thinning(X,nlags = 50):
autocorrelation = acf(X, nlags=nlags, fft=True)
thinning = np.argmin(np.abs(autocorrelation - 0.5)) + 1
return thinning, autocorrelation
def grad_log_lik(t):
a = np.sum(manual_grad(t[0],t[1],X),axis=0) - t[1]/SIGMA_2 -t[0]/SIGMA_1
return a
pvals = []
no_evals = []
for epsilon in np.linspace(0.001, 0.2,25):
THINNING_ESTIMAE = 10**4
sample,evals = austerity(vectorized_log_lik,log_density_prior, X,epsilon,batch_size=50, chain_size=THINNING_ESTIMAE, thinning=1, theta_t=np.random.randn(2))
thinning, autocorr = get_thinning(sample[:,0])
print(' - thinning for epsilon:',thinning,epsilon)
TEST_SIZE = 500
e_pvals = []
e_no_evals = []
for mc_reps in range(50):
print(mc_reps)
sample, evals = austerity(vectorized_log_lik,log_density_prior, X,epsilon,batch_size=50,chain_size=TEST_SIZE + MAGIC_BURNIN_NUMBER, thinning=thinning, theta_t=np.random.randn(2))
X = gen_X(SAMPLE_SIZE)
def vectorized_log_lik(X, theta):
return _vector_of_log_likelihoods(theta[0], theta[1], X)
def log_density_prior(theta):
return np.log(norm.pdf(theta[0], 0, SIGMA_1)) + np.log(
norm.pdf(theta[1], 0, SIGMA_2))
sample = austerity(vectorized_log_lik,
log_density_prior,
X,
0.01,
batch_size=50,
chain_size=20 * 1000,
thinning=1,
theta_t=np.random.randn(2))
print(acf(sample[:, 1], nlags=50))
#
# import seaborn as sns
# sns.set(color_codes=True)
# with sns.axes_style("white"):
# pr = sns.jointplot(x=sample[:,0], y=sample[:,1], kind="kde", color="k");
#
# # pr.savefig('../../write_up/img/mcmc_sample.pdf')
#
# sns.plt.show()