def calculate_kld(data_1_fn,data_2_fn,names,n_samples=2000):
assert isinstance(data_1_fn,str)
assert isinstance(data_2_fn,str)
assert isinstance(n_samples,int)
assert os.path.isfile(data_1_fn)
assert os.path.isfile(data_1_fn)
data_1 = PyposmatDataFile()
data_1.read(filename=data_1_fn)
data_2 = PyposmatDataFile()
data_2.read(filename=data_2_fn)
w1,v1 = linalg.eig(np.cov(data_1.df[names].T))
w2,v2 = linalg.eig(np.cov(data_2.df[names].T))
cov1_ill_conditioned = any([k < 0 for k in w1.tolist()])
cov2_ill_conditioned = any([k < 0 for k in w2.tolist()])
any_ill_conditioned = any([cov1_ill_conditioned,cov2_ill_conditioned])
if any_ill_conditioned:
print('using ill-conditioned kde')
kde_1 = GaussianKde(data_1.df[names].T)
print(kde_1.n, kde_1.d)
kde_2 = GaussianKde(data_2.df[names].T)
else:
kde_1 = gaussian_kde(data_1.df[names].T)
kde_2 = gaussian_kde(data_2.df[names].T)
kld = kullbach_lieber_divergence(kde_1,kde_2,n_samples)
return kld
def dev__kld_calculation_1d_kde():
n_samples_normal = 1000
n_samples_kde = 1000
rv_norm = norm(0,1)
X_norm = rv_norm.rvs(size=1000)
rv_kde_1 = gaussian_kde(X_norm)
X_kde = rv_kde_1.resample(size=1000)
rv_kde_2 = gaussian_kde(X_kde)
kld = kullbach_lieber_divergence(rv_kde_1,rv_kde_2,1000)
print(kld)
def test__kld_calculation_1d_kde():
n_samples_normal = 1000
n_samples_kde = 1000
rv_norm = norm(0,1)
X_norm = rv_norm.rvs(size=1000)
rv_kde_1 = gaussian_kde(X_norm)
X_kde = rv_kde_1.resample(size=1000)
rv_kde_2 = gaussian_kde(X_kde)
kld = kullbach_lieber_divergence(rv_kde_1,rv_kde_2,1000)
assert type(kld)==tuple
assert kld[0]>0
assert kld[0]>0
def dev__kld_calculation_1d_kde():
print(80*'-')
print('{:^80}'.format('dev__kld_calculation_1d_kde'))
print(80*'-')
n_samples_normal = 1000
n_samples_kde = 1000
rv_norm = norm(0,1)
X_norm = rv_norm.rvs(size=1000)
rv_kde_1 = gaussian_kde(X_norm)
X_kde = rv_kde_1.resample(size=1000)
rv_kde_2 = gaussian_kde(X_kde)
assert isinstance(rv_kde_1,gaussian_kde)
assert isinstance(rv_kde_1,gaussian_kde)
kld = kullbach_lieber_divergence(rv_kde_1,rv_kde_2,1000)
print(kld)
return kld
except FileNotFoundError as e:
print("the number of max iterations is actually {}".format(i - 1))
n_iterations = i - 1
break
# comparing kde estimates
print('compare kde estimates')
kld = [-1]
for i in range(1, n_iterations):
df_0 = data['kde'][i - 1].df
df_1 = data['kde'][i].df
df_0_p = df_0[config.parameter_names]
df_1_p = df_1[config.parameter_names]
nr0, nc0 = df_0_p.shape
nr1, nc1 = df_1_p.shape
#print('nrows:: {}={},{}={}'.format(i,nr0,i+1,nr1))
#print('ncols:: {}={},{}={}'.format(i,nc0,i+1,nc1))
X0 = df_0_p.values
X1 = df_1_p.values
#print('X0:',X0.shape,type(X0))
silverman86_h0 = Silverman1986_h(X0.T)
silverman86_h1 = Silverman1986_h(X1.T)
chiu99_h0 = Chiu1999_h(X0.T)
chiu99_h1 = Chiu1999_h(X1.T)
kde_0 = gaussian_kde(X0.T, chiu99_h0)
kde_1 = gaussian_kde(X1.T, chiu99_h1)
kld.append(kullbach_lieber_divergence(kde_0, kde_1, 400))
print(i, silverman86_h1, chiu99_h1, kld[i])
for i, v in enumerate(kld):
print(i, v)
A = A * n * np.eye(n)
return A
n_samples_normal = 1000
n_samples_kde = 1000
rv_norm = norm(0, 1)
X_norm = rv_norm.rvs(size=1000)
print('X_norm', X_norm.shape)
rv_kde_1 = gaussian_kde(X_norm)
X_kde = rv_kde_1.resample(size=1000)
print('X_kde', X_kde.shape)
rv_kde_2 = gaussian_kde(X_kde)
kld = kullbach_lieber_divergence(rv_kde_1, rv_kde_2, 1000)
print(kld)
exit()
import matplotlib.pyplot as plt
xmin = min(X_norm.min(), X_kde.min())
xmax = max(X_norm.max(), X_kde.max())
x = np.linspace(xmin, xmax, 1000)
fig, ax = plt.subplots()
ax.plot(x, rv_norm.pdf(x))
ax.plot(x, rv_kde.pdf(x))
plt.show()
configuration_fn = test_case['Si__sw']['configuration_fn']
configuration = PyposmatConfigurationFile()
configuration.read(filename=configuration_fn)
free_parameter_names = configuration.free_parameter_names
print('free_parameter_names:{}'.format(free_parameter_names))
for i in range(n_kde_files):
if i > 0:
print(80*'-')
print('i_iteration:{}'.format(i))
data_directory = test_case['Si__sw']['data_directory']
kde_file_fn_1 = os.path.join(data_directory,'pyposmat.kde.{}.out'.format(i))
kde_file_fn_2 = os.path.join(data_directory,'pyposmat.kde.{}.out'.format(i+1))
print('kde_file_fn_1:{}'.format(kde_file_fn_1))
print('kde_file_fn_2:{}'.format(kde_file_fn_2))
kde_file_1 = PyposmatDataFile()
kde_file_1.read(filename=kde_file_fn_1)
kde_rv_1 = gaussian_kde(kde_file_1.df[free_parameter_names].T)
kde_file_2 = PyposmatDataFile()
kde_file_2.read(filename=kde_file_fn_2)
kde_rv_2 = gaussian_kde(kde_file_2.df[free_parameter_names].T)
kld = kullbach_lieber_divergence(kde_rv_1,kde_rv_2,1000)
print(kld)
dev__kld_calculation_1d_kde()