def pdf_plot(fig2, indata, i, xbrs, ybrs, munits):
''' creates a Gaussian pdf plot of any array. If the array has multiple dimensions
it is flattened first. Relies on the Python "Statistics" and "Matplotlib" libraries.
http://bonsai.hgc.jp/~mdehoon/software/python/
http://matplotlib.org/'''
import statistics as st
import numpy as np
import matplotlib.pyplot as plt
from scipy import stats
colours=['Teal', 'MediumBlue', 'DodgerBlue', 'DarkTurquoise', 'Chartreuse', 'Green', 'Yellow', 'Red', 'Orange', 'Chocolate', 'DarkRed', 'Black']
pshape=np.zeros((12))
ploc=np.zeros((12))
pscale=np.zeros((12))
##flatten array if necessary
#indata=indata.flatten(1)
# adjust weight or bandwidth
#weight=[1]*len(z4)
#h1=bandwidth(z4,weight,"Epanechnikov")
#h2=bandwidth(z4,weight,"Gaussian")
# do pdf calculation
y0,x0=st.pdf(indata)
yg,xg=st.pdf(indata, kernel="Gaussian")
if i > 10:
plt.plot(x0,y0,color='black',linewidth=2.5)
plt.title(' ')
plt.title('1961-1990', fontsize=14)
else:
plt.plot(x0,y0,colours[i])
plt.title(' ')
plt.title('2061-2090', fontsize=14)
# limits axes
yy1=ybrs[0]
yy2=ybrs[1]
xx1=xbrs[0]
xx2=xbrs[1]
plt.ylim((yy1,yy2))
plt.xlim((xx1,xx2))
plt.locator_params(nbins=7)
# set axis label size
plt.rc('font', size=12)
#fig2.set_xlabel(munits, fontsize=11)
x=[5,25,50,75,95]
index_pdf = stats.norm.pdf(x)
pshape[i], ploc[i], pscale[i] = stats.lognorm.fit(indata, floc=0)
def kl_divergence_coalescent_waiting_times(allele_waiting_time_dist,
haploid_pop_size):
"""
`allele_branch_len_dist` is a dictionary with number of alleles as keys
and a list of waiting times associated with that number of alleles as
values. `haploid_pop_size` is the population size in terms of total numbers
of genes. This returns a the KL-divergence between the distribution of
waiting times and the Kingman coalescent distribution.
D_{\mathrm{KL}}(P\|Q) = \sum_i P(i) \log \frac{P(i)}{Q(i)}.
"""
d_kl = 0.0
for k, wts in allele_waiting_time_dist.items():
p = float(probability.binomial_coefficient(k,
2)) / haploid_pop_size
for t in wts:
# Kernel types:
#
# 'E' or 'Epanechnikov'
# Epanechnikov kernel (default)
#
# 'U' or 'Uniform'
# Uniform kernel
#
# 'T' or 'Triangle'
# Triangle kernel
#
# 'G' or 'Gaussian'
# Gaussian kernel
#
# 'B' or 'Biweight'
# Quartic/biweight kernel
#
# '3' or 'Triweight'
# Triweight kernel
#
# 'C' or 'Cosine'
# Cosine kernel
q = de_hoon_lib.pdf(wts, [k], kernel='Gaussian')
if q == 0:
q = 1e-100
d_kl += p * math.log(p / q)
return d_kl
def kl_divergence_coalescent_waiting_times(allele_waiting_time_dist, haploid_pop_size):
"""
`allele_branch_len_dist` is a dictionary with number of alleles as keys
and a list of waiting times associated with that number of alleles as
values. `haploid_pop_size` is the population size in terms of total numbers
of genes. This returns a the KL-divergence between the distribution of
waiting times and the Kingman coalescent distribution.
D_{\mathrm{KL}}(P\|Q) = \sum_i P(i) \log \frac{P(i)}{Q(i)}.
"""
d_kl = 0.0
for k, wts in allele_waiting_time_dist.items():
p = float(probability.binomial_coefficient(k, 2)) / haploid_pop_size
for t in wts:
# Kernel types:
#
# 'E' or 'Epanechnikov'
# Epanechnikov kernel (default)
#
# 'U' or 'Uniform'
# Uniform kernel
#
# 'T' or 'Triangle'
# Triangle kernel
#
# 'G' or 'Gaussian'
# Gaussian kernel
#
# 'B' or 'Biweight'
# Quartic/biweight kernel
#
# '3' or 'Triweight'
# Triweight kernel
#
# 'C' or 'Cosine'
# Cosine kernel
q = de_hoon_lib.pdf(wts, [k], kernel = 'Gaussian')
if q == 0:
q = 1e-100
d_kl += p * math.log(p/q)
return d_kl
def plothist(x, bin=None, nbins=None, xrange=None, yrange=None, min=None,
max=None, overplot=False, color='black', xlog=False, ylog=False,
nan=False, weights=None, norm=False, kernel=None, retpoints=False,
adaptive=False, adaptive_thresh=30, adaptive_depth=[2,10],
weight_norm=False, apply_func=None, **kw):
"""
Plot the 1D histogram
Example:
>> plothist(dat, bin=0.1, min=0, max=3)
Keyword parameters:
------------------
bin
the binsize(float)
nbins
number of bins(integer)
It cannot be specified together with the bin= parameter
xlog, ylog
log the appropriate axis
weights
the 1-D array of weights used in the histogram creation
nan
boolean flag to ignore nan's
norm
boolean flag to normalize the histogram by the peak value
min,max
range of data for which the histogram is constructed
retpoints
boolean parameter controlling whether to return or not the
computed histogram.
If yes the tuple with two arrays (bin centers, Number of points in bins)
is returned
overplot
boolean parameter for overplotting
adaptive
boolean for turning on/off the adaptive regime of
histogramming (adaptive bin size).
If True weights, nbins, bin,kernel parameters are ignored
adaptive_thresh
the limiting number of points in the bin for the adaptive
histogramming (default 30)
adaptive_depth
the list of two integers for the detalisation levels of
adaptive histogramming (default [2,10])
weight_norm
if True the value in each bin is mean weight of points within
the bin
"""
if nan:
ind = numpy.isfinite(x)
if weights is not None:
ind = numpy.isfinite(weights)&ind
dat = x[ind]
if weights is not None:
weights =weights[ind]
else:
dat = x
maxNone = False
if min is None:
min = numpy.min(dat)
if max is None:
maxNone = True
max = numpy.max(dat)
if bin is None and nbins is None:
nbins = 100
bin = (max - min) * 1. / nbins
elif nbins is None:
nbins = int(math.ceil((max - min) * 1. / bin))
if maxNone:
max = min + nbins * bin
elif bin is None:
bin = (max - min) * 1. / nbins
else:
warnings.warn("both bin= and nbins= keywords were specified in the plothist call",RuntimeWarning)
pass
# if both nbins and bin are defined I don't do anything
# it may be non-intuitive if kernel option is used, because
# it uses both nbins and bin options
if kernel is None:
if not adaptive:
if not np.isscalar(weights):
hh, loc = numpy.histogram(dat, range=(min, max), bins=nbins, weights=weights)
else:
hh, loc = numpy.histogram(dat, range=(min, max), bins=nbins)
hh = hh * weights
if weight_norm:
hh1, loc = numpy.histogram(dat, range=(min, max), bins=nbins, weights=None)
hh = hh*1./hh1
else:
import adabinner
hh, loc = adabinner.hist(dat, xmin=min, xmax=max, hi=adaptive_depth,
thresh=adaptive_thresh)
hh1 = np.repeat(hh,2)
loc1 = np.concatenate(([loc[0]],np.repeat(loc[1:-1],2),[loc[-1]]))
else:
loc1=numpy.linspace(min,max,nbins*5)
import statistics
if weights is not None:
hh1 = statistics.pdf( dat, loc1, h=bin/2.,kernel=kernel,weight=weights)*bin*len(dat)
else:
hh1 = statistics.pdf( dat, loc1, h=bin/2.,kernel=kernel)*bin*len(dat)
if overplot:
func = oplot
else:
func = plot
if norm:
hh1=hh1*1./hh1.max()
kw['ps'] = kw.get('ps') or 0
if 'yr' not in kw:
kw['yr']=[hh1.min(),hh1.max()]
if 'xr' not in kw:
kw['xr']=[min,max]
if apply_func is not None:
hh1 = apply_func (loc1,hh1)
func(loc1, hh1, color=color,
xlog=xlog, ylog=ylog, **kw)
if retpoints:
return 0.5*(loc[1:]+loc[:-1]),hh
def restrict(self,x,grid,domain):
edges = scipy.r_[domain[0],(grid[1:]+grid[:-1])/2.,domain[-1]]
# estimating the cumulative density to make the estimation conservative
rho = statistics.pdf(x,edges[1:-1],weight=weight,h=self.h)
return rho
def restrict(self, x, grid, domain):
edges = scipy.r_[domain[0], (grid[1:] + grid[:-1]) / 2., domain[-1]]
# estimating the cumulative density to make the estimation conservative
rho = statistics.pdf(x, edges[1:-1], weight=weight, h=self.h)
return rho
cb.set_label('log10(N)')
###################################################################################################
answer2=raw_input("Plot the magnitude distribution (Default=no)? (y/n) ")
if answer2=='y':
mpt.figure(1,figsize=(11,7),dpi=100)
mpt.suptitle("Magnitude Distribution")
mpt.subplot(211)
mpt.hexbin(a,mag,gridsize=200,bins='log',mincnt=1)
mpt.xlabel("$a'$ (AU)")
mpt.ylabel("$Magnitudes$")
mpt.subplot(212)
dist, mag_dist = stc.pdf(mag)
mpt.plot(mag_dist, dist,'r-',linewidth=2)
mpt.hist(mag,bins=200,normed=True)
mpt.xlim(3,20)
mpt.xlabel("$Magnitudes$")
mpt.ylabel("$f$")
# Completeza:
mpt.plot([17,17],[0,0.5],'b--')
mpt.plot([15.55,15.55],[0,0.5],'r--')
###################################################################################################
answer3=raw_input("Plot the taxonomic distribution (Default=no)? (y/n) ")
if answer3=='y':
def show_obs_sbr():
import numpy as np
import matplotlib.pyplot as plt
from copy import deepcopy
import statistics
import LetsgoSerializer as ls
from SparseBayes import SparseBayes
dimension = 1
# basisWidth = 0.05
# basisWidth = basisWidth**(1/dimension)
def dist_squared(X,Y):
import numpy as np
nx = X.shape[0]
ny = Y.shape[0]
return np.dot(np.atleast_2d(np.sum((X**2),1)).T,np.ones((1,ny))) + \
np.dot(np.ones((nx,1)),np.atleast_2d(np.sum((Y**2),1))) - 2*np.dot(X,Y.T);
def basis_func(X,basisWidth):
import numpy as np
C = X.copy()
BASIS = np.exp(-dist_squared(X,C)/(basisWidth**2))
return BASIS
def basis_vector(X,x,basisWidth):
import numpy as np
BASIS = np.exp(-dist_squared(x,X)/(basisWidth**2))
return BASIS
total_co_cs = None
total_inco_cs = None
for c in range(7):
co_cs = ls.load_model('_correct_confidence_score_class_%d.model'%c)
inco_cs = ls.load_model('_incorrect_confidence_score_class_%d.model'%c)
if total_co_cs == None:
total_co_cs = deepcopy(co_cs)
total_inco_cs = deepcopy(inco_cs)
else:
for k in co_cs.keys():
total_co_cs[k].extend(co_cs[k])
total_inco_cs[k].extend(inco_cs[k])
# plt.subplot(121)
title = {'multi':'Total of multiple actions',\
'multi2': 'Two actions',\
'multi3': 'Three actions',\
'multi4': 'Four actions',\
'multi5': 'Five actions',\
'total': 'Global',\
'yes': 'Affirm',\
'no': 'Deny',\
'bn': 'Bus number',\
'dp': 'Departure place',\
'ap': 'Arrival place',\
'tt': 'Travel time',\
'single': 'Total of single actions'
}
for k in total_co_cs.keys():
if not k in ['yes','no','bn','dp','ap','tt','multi2','multi3','multi4','multi5']:
continue
co = total_co_cs[k]
inco = total_inco_cs[k]
print 'length of correct: ',len(co)
print 'length of incorrect: ',len(inco)
# n,bins,patches = plt.hist([co,inco],bins=np.arange(0.0,1.1,0.1),\
# normed=0,color=['green','yellow'],\
# label=['Correct','Incorrect'],alpha=0.75)
try:
x_co = np.arange(0,1.001,0.001)
x_inco = np.arange(0,1.001,0.001)
h_co = statistics.bandwidth(np.array(co),weight=None,kernel='Gaussian')
print 'bandwidth of correct: ',h_co
# y_co,x_co = statistics.pdf(np.array(co),kernel='Gaussian',n=1000)
y_co = statistics.pdf(np.array(co),x=x_co,kernel='Gaussian')
print 'length of correct: ',len(x_co)
h_inco = statistics.bandwidth(np.array(inco),weight=None,kernel='Gaussian')
print 'bandwidth of incorrect: ',h_inco
# y_inco,x_inco = statistics.pdf(np.array(inco),kernel='Gaussian',n=1000)
y_inco = statistics.pdf(np.array(inco),x=x_inco,kernel='Gaussian')
print 'length of incorrect: ',len(x_inco)
y_co += 1e-10
y_inco = y_inco*(float(len(inco))/len(co)) + 1e-10
y_co_max = np.max(y_co)
print 'max of correct: ',y_co_max
y_inco_max = np.max(y_inco)
print 'max of incorrect: ',y_inco_max
y_max = max([y_co_max,y_inco_max])
print 'max of total: ',y_max
plt.plot(x_co,y_co/y_max,'g.-',alpha=0.75)
plt.plot(x_inco,y_inco/y_max,'r.-',alpha=0.75)
print x_co
print x_inco
y = y_co/(y_co + y_inco)
plt.plot(x_co,y,'b--',alpha=0.75)
m = SparseBayes()
X = np.atleast_2d(x_co).T
Y = np.atleast_2d(y).T
basisWidth=min([h_co,h_inco])
BASIS = basis_func(X,basisWidth)
try:
Relevant,Mu,Alpha,beta,update_count,add_count,delete_count,full_count = \
m.learn(X,Y,lambda x: basis_func(x,basisWidth))
ls.store_model({'data_points':X[Relevant],'weights':Mu,'basis_width':basisWidth},\
'_calibrated_confidence_score_sbr_%s.model'%k)
except RuntimeError as e:
print e
w_infer = np.zeros((BASIS.shape[1],1))
w_infer[Relevant] = Mu
Yh = np.dot(BASIS[:,Relevant],Mu)
e = Yh - Y
ED = np.dot(e.T,e)
print 'ED: %f'%ED
print np.dot(basis_vector(X[Relevant],np.ones((1,1))/2,basisWidth),Mu)
plt.plot(X.ravel(),Yh.ravel(),'yo-',alpha=0.75)
# plt.legend(loc='upper center')
plt.xlabel('Confidence Score')
plt.ylabel('Count')
plt.title(title[k])
# if k == 'multi5':
# plt.axis([0,1,0,1.2])
# elif k == 'multi4':
# plt.axis([0,1,0,10])
plt.grid(True)
plt.savefig(title[k]+'.png')
# plt.show()
plt.clf()
except (ValueError,RuntimeError) as e:
print e
