def compute(H, K, cdf=False):
H, K = Divergence.compute(H, K)
try:
if cdf:
if not (analytics.isValidCDF(H) and analytics.isValidCDF(K)):
raise smp.InvalidProbabilityDistributionException(
'Invalid cumulative distribution')
if not cdf:
if not (analytics.isValidPDF(H) and analytics.isValidPDF(K)):
raise smp.InvalidProbabilityDensityException(
'Invalid probability density')
#convert to cdf
H = analytics.pdfTocdf(H)
K = analytics.pdfTocdf(K)
all_keys = dict(H.items() + K.items()).keys()
analytics.fill_gaps(H, all_keys, cdf=True)
analytics.fill_gaps(K, all_keys, cdf=True)
max_diff = 0
#max_key=-1
for key in sorted(all_keys):
if math.fabs(H[key] - K[key]) > max_diff:
# max_key=key
max_diff = math.fabs(H[key] - K[key])
return max_diff #, max_key, H[max_key],K[max_key]
except smp.InvalidProbabilityDistributionException:
logging.exception('Invalid probability distribution')
except smp.InvalidProbabilityDensityException:
logging.exception('Invalid probability density')
def compute(H, K):
Divergence.compute(H, K)
try:
if analytics.isValidPDF(H) and analytics.isValidPDF(K):
all_keys = dict(H.items() + K.items()).keys()
analytics.fill_gaps(H, all_keys)
analytics.fill_gaps(K, all_keys)
total = 0
for key in sorted(all_keys):
if H[key] != 0:
if K[key] == 0:
raise ValueError(
'KL Divergence is not defined for the input. Please see the definition of KL-Divergence'
)
else:
total += H[key] * math.log(H[key] / K[key], 2)
return total
else:
raise smp.InvalidProbabilityDensityException(
'Invalid probability density')
except ValueError:
logging.error(
'KL Divergence is not defined for the input. Please see the definition of KL-Divergence'
)
except smp.InvalidProbabilityDensityException:
logging.exception('Invalid probability density')
def compute(H,K, cdf=False):
H,K=Divergence.compute(H, K)
try:
if cdf:
if not (analytics.isValidCDF(H) and analytics.isValidCDF(K)):
raise smp.InvalidProbabilityDistributionException('Invalid cumulative distribution')
if not cdf:
if not (analytics.isValidPDF(H) and analytics.isValidPDF(K)):
raise smp.InvalidProbabilityDensityException('Invalid probability density')
#convert to cdf
H=analytics.pdfTocdf(H)
K=analytics.pdfTocdf(K)
all_keys= dict(H.items()+K.items()).keys()
analytics.fill_gaps(H, all_keys,cdf=True)
analytics.fill_gaps(K, all_keys,cdf=True)
max_diff=0
#max_key=-1
for key in sorted(all_keys):
if math.fabs(H[key]-K[key])>max_diff:
# max_key=key
max_diff=math.fabs(H[key]-K[key])
return max_diff#, max_key, H[max_key],K[max_key]
except smp.InvalidProbabilityDistributionException:
logging.exception('Invalid probability distribution')
except smp.InvalidProbabilityDensityException:
logging.exception('Invalid probability density')
def distribution_save(H, K, file_name, cdf=False):
H,K=Divergence.compute(H, K)
try:
if cdf:
if not (analytics.isValidCDF(H) and analytics.isValidCDF(K)):
raise smp.InvalidProbabilityDistributionException('Invalid cumulative distribution')
if not cdf:
if not (analytics.isValidPDF(H) and analytics.isValidPDF(K)):
raise smp.InvalidProbabilityDensityException('Invalid probability density')
#convert to cdf
H=analytics.pdfTocdf(H)
K=analytics.pdfTocdf(K)
all_keys= dict(H.items()+K.items()).keys()
analytics.fill_gaps(H, all_keys, cdf=True)
analytics.fill_gaps(K, all_keys, cdf=True)
fileObject = open(file_name, 'w')
for key in sorted(all_keys):
fileObject.write(str(key)+' '+ str(K[key])+'\n')
# print("dict[%s] =" % key,K[key])
fileObject.close()
#==============================================================================
# plot_keys=[]
# plot_values=[]
# fileObject = open(file_name, 'w')
# for key in sorted(all_keys):
# plot_keys.append(key)
# plot_values.append(K[key])
#
# fileObject.write(str(key)+' '+ str(H[key])+'\n')
# print("dict[%s] =" % key,H[key])
# fileObject.close()
#
# plt.figure(1)
# plt.plot(plot_keys, plot_values, 'r-')
# plt.xscale('log')
# plt.yscale('log')
# plt.ylim([0,1])
# plt.ylabel('CDF')
# plt.xlabel('Degree')
# plt.savefig(file_name+ '_cdf.pdf')
# plt.clf()
#==============================================================================
except smp.InvalidProbabilityDistributionException:
logging.exception('Invalid probability distribution')
except smp.InvalidProbabilityDensityException:
logging.exception('Invalid probability density')
def compute(H,K,lamda):
try:
H,K=Divergence.compute(H, K)
if lamda <=0 or lamda >=1 :
raise ValueError('Invalid lambda value for LambdaDivergence. It should be in (0,1).')
if analytics.isValidPDF(H) and analytics.isValidPDF(K):
smoothH,_=analytics.smooth(H, K, lamda)
return lamda* KLDivergence.compute(H, smoothH)+ (1-lamda)* KLDivergence.compute(K, smoothH)
else:
print K
raise smp.InvalidProbabilityDensityException('Invalid probability density')
except smp.InvalidProbabilityDensityException:
logging.exception('Invalid probability density')
except ValueError:
logging.exception('Invalid lambda value for LambdaDivergence. It should be positive.')
def plot_single_distribution(normalized_distr, cdf=True, xlabel=None, title=None, *args, **kwds):
axis= plt.gca()
if xlabel is not None:
axis.set_xlabel(xlabel)
if title is not None:
axis.set_title(title)
if not cdf and not analytics.isValidPDF(normalized_distr):
raise sampling.InvalidProbabilityDensityException('Invalid probability density')
axis.set_ylim(0,1)
if cdf:
axis.set_ylabel('CDF')
if kwds.has_key('kind'):
fill_gaps(normalized_distr, range(min(normalized_distr.keys()), max(normalized_distr.keys())), cdf=True)
cdf=analytics.pdfTocdf(normalized_distr)
if analytics.trim_cdf(cdf):
Series(cdf,copy=True).plot(ax=axis, *args, **kwds)
else:
axis.set_ylabel('PDF')
if kwds.has_key('kind'):
fill_gaps(normalized_distr, range(min(normalized_distr.keys()), max(normalized_distr.keys())+1), cdf=False)
if analytics.trim_pdf(normalized_distr):
Series(normalized_distr,copy=True).plot(ax=axis, *args, **kwds)
xmin, xmax = plt.xlim()
axis.set_xlim(xmin-0.10*(xmax-xmin),xmax+0.05*(xmax-xmin))
def compute(H, K, lamda):
try:
H, K = Divergence.compute(H, K)
if lamda <= 0 or lamda >= 1:
raise ValueError(
'Invalid lambda value for LambdaDivergence. It should be in (0,1).'
)
if analytics.isValidPDF(H) and analytics.isValidPDF(K):
smoothH, _ = analytics.smooth(H, K, lamda)
return lamda * KLDivergence.compute(H, smoothH) + (
1 - lamda) * KLDivergence.compute(K, smoothH)
else:
print K
raise smp.InvalidProbabilityDensityException(
'Invalid probability density')
except smp.InvalidProbabilityDensityException:
logging.exception('Invalid probability density')
except ValueError:
logging.exception(
'Invalid lambda value for LambdaDivergence. It should be positive.'
)
def compute(H, K):
Divergence.compute(H, K)
try:
if analytics.isValidPDF(H) and analytics.isValidPDF(K):
all_keys= dict(H.items()+K.items()).keys()
analytics.fill_gaps(H,all_keys)
analytics.fill_gaps(K,all_keys)
total=0
for key in sorted(all_keys):
if H[key] != 0:
if K[key] == 0:
raise ValueError('KL Divergence is not defined for the input. Please see the definition of KL-Divergence')
else:
total+=H[key]*math.log(H[key]/K[key],2)
return total
else:
raise smp.InvalidProbabilityDensityException('Invalid probability density')
except ValueError:
logging.error('KL Divergence is not defined for the input. Please see the definition of KL-Divergence')
except smp.InvalidProbabilityDensityException:
logging.exception('Invalid probability density')
def plot_single_distribution(normalized_distr,
cdf=True,
xlabel=None,
title=None,
*args,
**kwds):
axis = plt.gca()
if xlabel is not None:
axis.set_xlabel(xlabel)
if title is not None:
axis.set_title(title)
if not cdf and not analytics.isValidPDF(normalized_distr):
raise sampling.InvalidProbabilityDensityException(
'Invalid probability density')
axis.set_ylim(0, 1)
if cdf:
axis.set_ylabel('CDF')
if kwds.has_key('kind'):
fill_gaps(normalized_distr,
range(min(normalized_distr.keys()),
max(normalized_distr.keys())),
cdf=True)
cdf = analytics.pdfTocdf(normalized_distr)
if analytics.trim_cdf(cdf):
Series(cdf, copy=True).plot(ax=axis, *args, **kwds)
else:
axis.set_ylabel('PDF')
if kwds.has_key('kind'):
fill_gaps(normalized_distr,
range(min(normalized_distr.keys()),
max(normalized_distr.keys()) + 1),
cdf=False)
if analytics.trim_pdf(normalized_distr):
Series(normalized_distr, copy=True).plot(ax=axis, *args, **kwds)
xmin, xmax = plt.xlim()
axis.set_xlim(xmin - 0.10 * (xmax - xmin), xmax + 0.05 * (xmax - xmin))