def __init__(self, refsample, refweights=None, ref_label='Reference'):
"""refsample is the reference (data) sample to be compared to."""
self.ref = np.asarray(refsample.ravel())
if refweights is None:
self.weights = None
else:
self.weights = np.asarray(refweights.ravel())
self.ref_mean = np.ma.average(self.ref, weights=self.weights)
self.ref_STD = weighted_std(self.ref, weights=self.weights)
self.ref_label = ref_label
def __init__(self, refsample, refweights=None, ref_label='Reference'):
"""refsample is the reference (data) sample to be compared to."""
self.ref = np.asarray(refsample.ravel())
if refweights is None:
self.weights = None
else:
self.weights = np.asarray(refweights.ravel())
self.ref_mean = np.ma.average(self.ref, weights=self.weights)
self.ref_STD = weighted_std(self.ref, weights=self.weights)
self.ref_label = ref_label
def get_coords(self, name, sample, weights=None, print_bool=False):
"""Computes theta=arccos(correlation),rad=stddev of sample
wrt. reference sample."""
# flatten
my_sample = sample.ravel()
if weights is None:
my_weights = weights
else:
my_weights = weights.ravel()
# (weighted) average and standard deviation of the sample
ave = np.ma.average(my_sample, weights=my_weights)
std = weighted_std(my_sample, weights=my_weights)
# (weighted) correlation coefficient of the sample with the reference
# after http://wapedia.mobi/en/Pearson_product-moment_correlation_coefficient?p=1
# NO WEIGHTS: corr = np.corrcoef(self.ref, my_sample) # [[1,rho],[rho,1]]
if my_weights is None:
my_weights = np.ones_like(self.ref)
cov_x_y = ( np.ma.sum( my_weights
* (my_sample - ave)
* (self.ref - self.ref_mean) )
/ np.ma.sum( my_weights ) )
cov_x_x = ( np.ma.sum( my_weights
* (my_sample - ave)**2)
/ np.ma.sum( my_weights ) )
cov_y_y = ( np.ma.sum( my_weights
* (self.ref - self.ref_mean)**2 )
/ np.ma.sum( my_weights ) )
corr = 1. * cov_x_y / (cov_x_x * cov_y_y)**0.5
theta = np.arccos(corr)
## # info to see how much does corr coeff change when use weighted corr vs non-weighted corr
## print '#' * 80
## mtxt = "corr WITHOUT weights = %.2f %%, corr WITH weights = %.2f %%, abs diff = %.2f %%, rel diff = %.2f %%, "
## no_weight_corr = np.corrcoef(self.ref, my_sample)[0,1]
## mtup = (100. * no_weight_corr,
## 100. * corr,
## 100. * abs(corr - no_weight_corr),
## 100. * 2 * abs(corr - no_weight_corr) / (corr + no_weight_corr)
## )
## print mtxt %mtup
## print '#' * 80
## #
if print_bool:
print "std=%.2f and corr=%.2f"%(std, corr[0,1]), "for %s"%name
return theta, std, corr
def get_coords(self, name, sample, weights=None, print_bool=False):
"""Computes theta=arccos(correlation),rad=stddev of sample
wrt. reference sample."""
# flatten
my_sample = sample.ravel()
if weights is None:
my_weights = weights
else:
my_weights = weights.ravel()
# (weighted) average and standard deviation of the sample
ave = np.ma.average(my_sample, weights=my_weights)
std = weighted_std(my_sample, weights=my_weights)
# (weighted) correlation coefficient of the sample with the reference
# after http://wapedia.mobi/en/Pearson_product-moment_correlation_coefficient?p=1
# NO WEIGHTS: corr = np.corrcoef(self.ref, my_sample) # [[1,rho],[rho,1]]
if my_weights is None:
my_weights = np.ones_like(self.ref)
cov_x_y = (np.ma.sum(my_weights * (my_sample - ave) *
(self.ref - self.ref_mean)) /
np.ma.sum(my_weights))
cov_x_x = (np.ma.sum(my_weights * (my_sample - ave)**2) /
np.ma.sum(my_weights))
cov_y_y = (np.ma.sum(my_weights * (self.ref - self.ref_mean)**2) /
np.ma.sum(my_weights))
corr = 1. * cov_x_y / (cov_x_x * cov_y_y)**0.5
theta = np.arccos(corr)
## # info to see how much does corr coeff change when use weighted corr vs non-weighted corr
## print '#' * 80
## mtxt = "corr WITHOUT weights = %.2f %%, corr WITH weights = %.2f %%, abs diff = %.2f %%, rel diff = %.2f %%, "
## no_weight_corr = np.corrcoef(self.ref, my_sample)[0,1]
## mtup = (100. * no_weight_corr,
## 100. * corr,
## 100. * abs(corr - no_weight_corr),
## 100. * 2 * abs(corr - no_weight_corr) / (corr + no_weight_corr)
## )
## print mtxt %mtup
## print '#' * 80
## #
if print_bool:
print "std=%.2f and corr=%.2f" % (std, corr[0, 1]), "for %s" % name
return theta, std, corr
def check_sample(self, sample, weights=None, threshold=1e-12):
"""Check for the sample if the following relation holds:
RMS^2 - STD^2 - STD_ref^2 + 2*STD*STD_ref*COR < threshold.
"""
my_sample = sample.ravel()
if weights is None:
my_weights = weights
else:
my_weights = weights.ravel()
means = np.ma.average(my_sample, weights=my_weights)
STDs = weighted_std(my_sample, weights=my_weights)
terms = ((self.ref - self.ref_mean) - (my_sample - means))**2
RMSs = (1. * np.sum(terms) / np.size(self.ref))**0.5
# (weighted) correlation coefficient of the sample with the reference
# after http://wapedia.mobi/en/Pearson_product-moment_correlation_coefficient?p=1
# NO WEIGHTS: CORs = np.corrcoef(self.ref, my_sample)[0,1] # [[1,rho],[rho,1]]
if my_weights is None:
my_weights = 1.
cov_x_y = ( np.ma.sum( my_weights
* (my_sample - means)
* (self.ref - self.ref_mean) )
/ np.ma.sum( my_weights ) )
cov_x_x = ( np.ma.sum( my_weights
* (my_sample - means)**2)
/ np.ma.sum( my_weights ) )
cov_y_y = ( np.ma.sum( my_weights
* (self.ref - self.ref_mean)**2 )
/ np.ma.sum( my_weights ) )
CORs = 1. * cov_x_y / (cov_x_x * cov_y_y)**0.5
value = abs(RMSs - (STDs**2 + self.ref_STD**2 - 2*STDs*self.ref_STD*CORs)**0.5)
test = value < threshold
## a = RMSs
## b = (STDs**2 + self.ref_STD**2 - 2*STDs*self.ref_STD*CORs)**0.5
## value1 = abs(2. * (a - b) / (a + b))
## value2 = abs(a - b)
## print "BITE, val = %.2e, val1 = %.2e, val2 = %.2e" %(value, value1, value2)
return test, value
def check_sample(self, sample, weights=None, threshold=1e-12):
"""Check for the sample if the following relation holds:
RMS^2 - STD^2 - STD_ref^2 + 2*STD*STD_ref*COR < threshold.
"""
my_sample = sample.ravel()
if weights is None:
my_weights = weights
else:
my_weights = weights.ravel()
means = np.ma.average(my_sample, weights=my_weights)
STDs = weighted_std(my_sample, weights=my_weights)
terms = ((self.ref - self.ref_mean) - (my_sample - means))**2
RMSs = (1. * np.sum(terms) / np.size(self.ref))**0.5
# (weighted) correlation coefficient of the sample with the reference
# after http://wapedia.mobi/en/Pearson_product-moment_correlation_coefficient?p=1
# NO WEIGHTS: CORs = np.corrcoef(self.ref, my_sample)[0,1] # [[1,rho],[rho,1]]
if my_weights is None:
my_weights = 1.
cov_x_y = (np.ma.sum(my_weights * (my_sample - means) *
(self.ref - self.ref_mean)) /
np.ma.sum(my_weights))
cov_x_x = (np.ma.sum(my_weights * (my_sample - means)**2) /
np.ma.sum(my_weights))
cov_y_y = (np.ma.sum(my_weights * (self.ref - self.ref_mean)**2) /
np.ma.sum(my_weights))
CORs = 1. * cov_x_y / (cov_x_x * cov_y_y)**0.5
value = abs(RMSs - (STDs**2 + self.ref_STD**2 -
2 * STDs * self.ref_STD * CORs)**0.5)
test = value < threshold
## a = RMSs
## b = (STDs**2 + self.ref_STD**2 - 2*STDs*self.ref_STD*CORs)**0.5
## value1 = abs(2. * (a - b) / (a + b))
## value2 = abs(a - b)
## print "BITE, val = %.2e, val1 = %.2e, val2 = %.2e" %(value, value1, value2)
return test, value
