def plotSTi(self, width=0.5, addval=True, sortit=True, *args, **kwargs):
'''
Plot a barchart of the given STi
Parameters
-----------
width : float (0-1)
width of the bars in the barchart
addval : bool
if True, the morris mu values are added to the graph
sortit : bool
if True, larger values (in absolute value) are plotted closest to
the y-axis
*args, **kwargs : args
passed to the matplotlib.bar
'''
if self.repl > 1:
print 'Table generates only output of first line!'
fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1 = plotbar(ax1,
self.STi[0],
self._namelist,
width=width,
addval=addval,
sortit=True,
*args,
**kwargs)
ax1.set_ylabel(r'$S_Ti$', fontsize=20)
ax1.yaxis.label.set_rotation('horizontal')
ax1.yaxis.set_label_coords(-0.02, 1.)
plt.draw()
return fig, ax1
def plotSTi(self, width = 0.5, addval = True, sortit = True,
*args, **kwargs):
'''
Plot a barchart of the given STi
Parameters
-----------
width : float (0-1)
width of the bars in the barchart
addval : bool
if True, the morris mu values are added to the graph
sortit : bool
if True, larger values (in absolute value) are plotted closest to
the y-axis
*args, **kwargs : args
passed to the matplotlib.bar
'''
if self.repl > 1:
print 'Table generates only output of first line!'
fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1 = plotbar(ax1, self.STi[0], self._namelist, width = width,
addval = addval, sortit = True, *args, **kwargs)
ax1.set_ylabel(r'$S_Ti$',fontsize=20)
ax1.yaxis.label.set_rotation('horizontal')
ax1.yaxis.set_label_coords(-0.02, 1.)
plt.draw()
return fig, ax1
def plotsens(self, indice='PE', width = 0.5, addval = True, sortit = True, outputid = 0,
*args, **kwargs):
'''
Plot a barchart of either CAS, CTRS or PE based Senstivitity
Parameters
-----------
indice: PE, CAS or CTRS
choose which indice you want to show
width : float (0-1)
width of the bars in the barchart
addval : bool
if True, the sensitivity values are added to the graph
sortit : bool
if True, larger values (in absolute value) are plotted closest to
the y-axis
outputid : int
the output to use whe multiple are compared; starts with 0
*args, **kwargs : args
passed to the matplotlib.bar
Returns
----------
fig : matplotlib.figure.Figure object
figure containing the output
ax1 : axes.AxesSubplot object
the subplot
'''
if indice == 'PE':
indtoplot = self.PE_SENS[:,outputid]
elif indice == 'CAS':
indtoplot = self.CAS_SENS[:,outputid]
elif indice == 'CTRS':
indtoplot = self.CTRS_SENS[:,outputid]
else:
raise Exception('Choose either PE, CAS or CTRS')
fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1 = plotbar(ax1, indtoplot, self._namelist, width = width,
addval = addval, sortit = True, *args, **kwargs)
ax1.set_ylabel(indice,fontsize=20)
ax1.yaxis.label.set_rotation('horizontal')
ax1.yaxis.set_label_coords(-0.02, 1.)
plt.draw()
return fig, ax1
def plotmustar(self, width = 0.5, addval = True, sortit = True, outputid = 0,
*args, **kwargs):
'''
Plot a barchart of the given mustar
mu* is a measure for the first-order effect on the model output.
By taking the average of the absolute values of the parameter
distribution, the absolute effect on the output can be calculated. This
is very useful when you are working with non-monotonic functions. The
drawback is that you lose information about the direction of influence
(is this factor influencing the output in a positive or negative way?).
For the direction of influence use plotmustar!
Parameters
-----------
width : float (0-1)
width of the bars in the barchart
addval : bool
if True, the morris values are added to the graph
sortit : bool
if True, larger values (in absolute value) are plotted closest to
the y-axis
outputid : int
the output to use whe multiple are compared; starts with 0
*args, **kwargs : args
passed to the matplotlib.bar; width is already used
Returns
----------
fig: matplotlib.figure.Figure object
figure containing the output
ax1: axes.AxesSubplot object
the subplot
'''
fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1 = plotbar(ax1, self.mustar[outputid*self._ndim:(outputid+1)*self._ndim], self._namelist, width = width,
addval = addval, sortit = True, *args, **kwargs)
ax1.set_ylabel(r'$\mu*$',fontsize=20)
ax1.yaxis.label.set_rotation('horizontal')
ax1.yaxis.set_label_coords(-0.02, 1.)
plt.draw()
return fig, ax1
def plotmu(self, width = 0.5, addval = True, sortit = True, outputid = 0,
*args, **kwargs):
'''
Plot a barchart of the given mu
mu is a measure for the first-order effect on the model output. However
the use of mu can be tricky because if the model is non-monotonic
negative elements can be in the parameter distribution and by taking the
mean of the variance (= mu!) the netto effect is cancelled out! Should
always be used together with plotsigma in order to see whether higher
order effects are occuring, high sigma values with low mu values can
be caused by non-monotonicity of functions. In order to check this use
plotmustar and/or plotmustarsigma
Parameters
-----------
width : float (0-1)
width of the bars in the barchart
addval : bool
if True, the morris mu values are added to the graph
sortit : bool
if True, larger values (in absolute value) are plotted closest to
the y-axis
outputid : int
the output to use whe multiple are compared; starts with 0
*args, **kwargs : args
passed to the matplotlib.bar
Returns
----------
fig : matplotlib.figure.Figure object
figure containing the output
ax1 : axes.AxesSubplot object
the subplot
'''
fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1 = plotbar(ax1, self.mu[outputid*self._ndim:(outputid+1)*self._ndim], self._namelist, width = width,
addval = addval, sortit = True, *args, **kwargs)
ax1.set_ylabel(r'$\mu$',fontsize=20)
ax1.yaxis.label.set_rotation('horizontal')
ax1.yaxis.set_label_coords(-0.02, 1.)
plt.draw()
return fig, ax1
def plotsens(self, indice="PE", width=0.5, addval=True, sortit=True, outputid=0, *args, **kwargs):
"""
Plot a barchart of either CAS, CTRS or PE based Senstivitity
Parameters
-----------
indice: PE, CAS or CTRS
choose which indice you want to show
width : float (0-1)
width of the bars in the barchart
addval : bool
if True, the sensitivity values are added to the graph
sortit : bool
if True, larger values (in absolute value) are plotted closest to
the y-axis
outputid : int
the output to use whe multiple are compared; starts with 0
*args, **kwargs : args
passed to the matplotlib.bar
Returns
----------
fig : matplotlib.figure.Figure object
figure containing the output
ax1 : axes.AxesSubplot object
the subplot
"""
if indice == "PE":
indtoplot = self.PE_SENS[:, outputid]
elif indice == "CAS":
indtoplot = self.CAS_SENS[:, outputid]
elif indice == "CTRS":
indtoplot = self.CTRS_SENS[:, outputid]
else:
raise Exception("Choose either PE, CAS or CTRS")
fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1 = plotbar(ax1, indtoplot, self._namelist, width=width, addval=addval, sortit=True, *args, **kwargs)
ax1.set_ylabel(indice, fontsize=20)
ax1.yaxis.label.set_rotation("horizontal")
ax1.yaxis.set_label_coords(-0.02, 1.0)
plt.draw()
return fig, ax1
def plotsigma(self, width = 0.5, addval = True, sortit = True, outputid = 0,
*args, **kwargs):
'''
Plot a barchart of the given sigma
sigma is a measure for the higher order effects (e.g. curvatures and
interactions). High sigma values typically occurs when for example
parameters are correlated, the model is non linear,... The sum of all
these effects is sigma. For linear models without any correlation sigma
should be approximately zero.
Parameters
-----------
width : float (0-1)
width of the bars in the barchart
addval : bool
if True, the morris values are added to the graph
sortit : bool
if True, larger values (in absolute value) are plotted closest to
the y-axis
outputid : int
the output to use whe multiple are compared; starts with 0
*args, **kwargs : args
passed to the matplotlib.bar
Returns
----------
fig : matplotlib.figure.Figure object
figure containing the output
ax1 : axes.AxesSubplot object
the subplot
'''
fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1 = plotbar(ax1, self.sigma[outputid*self._ndim:(outputid+1)*self._ndim], self._namelist, width = width,
addval = addval, sortit = True, *args, **kwargs)
ax1.set_ylabel(r'$\sigma$',fontsize=20)
ax1.yaxis.label.set_rotation('horizontal')
ax1.yaxis.set_label_coords(-0.02, 1.)
plt.draw()
return fig, ax1
def runTestModel(self, ai):
''' Run a testmodel
Use a testmodel to get familiar with the method and try things out. The
implemented model is the G Sobol function: testfunction with
analytical solution, moire information, see [M3]_
Parameters
-----------
ai : list
list with the input factors (equal size as number of factors)
Examples
---------
>>> ai=[78, 12, 0.5, 2, 97, 33]
>>> Xi = [(0.0,1.0,r'$X_1$'),(0.0,1.0,r'$X_2$'),(0.0,1.0,r'$X_3$'),
(0.0,1.0,r'$X_4$'),(0.0,1.0,r'$X_5$'),(0.0,1.0,r'$X_6$')]
>>> #set up the morris screening class
>>> sm = MorrisScreening(Xi,ModelType = 'testmodel')
>>> #Get an optimized set of trajectories
>>> OptMatrix, OptOutVec = sm.Optimized_Groups(nbaseruns=100,
intervals = 4,
noptimized=4,
Delta = 0.4)
>>> #compare the selected trajects with the general
>>> sm.Optimized_diagnostic(width=0.15)
>>> #run a testmodel and get the outputs
>>> sm.runTestModel(ai)
'''
output = np.zeros((sm.OptOutMatrix.shape[0],1))
for i in range(sm.OptOutMatrix.shape[0]):
output[i,:] = analgfunc(ai,sm.OptOutMatrix[i,:])
SAmeas_out, OutMatrix = self.Morris_Measure_Groups(output)
print 'Higher values of ai correspond to lower importance of Xi \n'
#Analytical to compare -> G
Vi = np.zeros(len(ai))
for i in range(len(ai)):
Vi[i] = (1./3.)/(1.+ai[i])**2
VTi = np.zeros(len(ai))
for i in range(len(ai)):
Vj = np.prod(np.delete(Vi,i)+1.)
VTi[i] = Vi[i] * Vj
Vtot = 1.
for i in range(len(ai)):
Vtot = Vtot * (1+Vi[i])
Vtot = Vtot -1.
print 'Morris gives only qualitive measures of importance, \n'
print 'a correspondance between STi and mustar is expected \n'
print 'and compared here \n'
print ' \n'
print 'Analytical Solution for STi: \n'
print VTi/Vtot
print 'The Morris mu* results: \n'
print self.mustar
print 'A barplot is generated...'
fig, ax1 = self.plotmustar(width=0.15, ec='grey',fc='grey')
ax1.set_title('Morris screening result')
fig2 = plt.figure()
ax2 = fig2.add_subplot(111)
ax2 = plotbar(ax2, VTi/Vtot, self._namelist, width = 0.15,
addval = True, sortit = True, ec='grey',fc='grey')
ax2.set_title('Analytical result')
def plot_SRC(self, width = 0.2, addval = True, sortit = True, outputid = 'all',
ncols = 3, outputnames=[], *args, **kwargs):
'''
Plot a barchart of the SRC values; actually a Tornadoplot in the
horizontal direction. More in the style of the other methods.
TODO: make for subset of outputs, also in other methods; currently all or 1
Parameters
-----------
width : float (0-1)
width of the bars in the barchart
addval : bool
if True, the sensitivity values are added to the graph
sortit : bool
if True, larger values (in absolute value) are plotted closest to
the y-axis
outputid : int or all
the output to use whe multiple are compared; starts with 0
if all, the different outputs are plotted in subplots
ncols : int
number of columns to put the subplots in
outputnames : list of strings
[] to plot no outputnames, otherwise list of strings equal to the
number of outputs
*args, **kwargs : args
passed to the matplotlib.bar
Returns
----------
fig : matplotlib.figure.Figure object
figure containing the output
ax1 : axes.AxesSubplot object
the subplot
Examples
---------
>>> #plot single SRC
>>> reg.plot_SRC(outputid=0, width=0.3, sortit = True,
ec='grey',fc='grey')
>>> #plot single SRC (assume 4 outputs)
>>> reg.plot_SRC(outputid = 'all' , ncols = 2,
outputnames=['o1','o2','o3','o4'], ec='grey', fc='grey')
'''
if outputid == 'all':
#define number of rows
numsrc = self.SRC.shape[1]
nrows = np.divide(numsrc,ncols)
if np.remainder(numsrc,ncols)>0:
nrows+=1
#prepare plots
fig, axes = plt.subplots(nrows = nrows, ncols = ncols, figsize=(40,40), facecolor = 'white')
fig.subplots_adjust(hspace=0.25, wspace=0.2)
i=0
for ax in axes.flat:
if i<numsrc:
ax = plotbar(ax, self.SRC[:,i], self._namelist,
width = width, addval = addval,
sortit = True, *args, **kwargs)
ax.set_ylabel('SRC',fontsize=15)
ax.yaxis.label.set_rotation('horizontal')
ax.yaxis.set_label_coords(-0.02, 1.)
if outputnames:
ax.set_title(outputnames[i])
else:
ax.set_axis_off()
i+=1
else:
print 'SRC values of output ',outputid,' is shown in graph'
fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1 = plotbar(ax1, self.SRC[:,outputid], self._namelist, width = width,
addval = addval, sortit = True, *args, **kwargs)
ax1.set_ylabel('SRC',fontsize=20)
ax1.yaxis.label.set_rotation('horizontal')
ax1.yaxis.set_label_coords(-0.02, 1.)
plt.draw()
return fig, ax1
def plot_SRC(self,
width=0.2,
addval=True,
sortit=True,
outputid='all',
ncols=3,
outputnames=[],
*args,
**kwargs):
'''
Plot a barchart of the SRC values; actually a Tornadoplot in the
horizontal direction. More in the style of the other methods.
TODO: make for subset of outputs, also in other methods; currently all or 1
Parameters
-----------
width : float (0-1)
width of the bars in the barchart
addval : bool
if True, the sensitivity values are added to the graph
sortit : bool
if True, larger values (in absolute value) are plotted closest to
the y-axis
outputid : int or all
the output to use whe multiple are compared; starts with 0
if all, the different outputs are plotted in subplots
ncols : int
number of columns to put the subplots in
outputnames : list of strings
[] to plot no outputnames, otherwise list of strings equal to the
number of outputs
*args, **kwargs : args
passed to the matplotlib.bar
Returns
----------
fig : matplotlib.figure.Figure object
figure containing the output
ax1 : axes.AxesSubplot object
the subplot
Examples
---------
>>> #plot single SRC
>>> reg.plot_SRC(outputid=0, width=0.3, sortit = True,
ec='grey',fc='grey')
>>> #plot single SRC (assume 4 outputs)
>>> reg.plot_SRC(outputid = 'all' , ncols = 2,
outputnames=['o1','o2','o3','o4'], ec='grey', fc='grey')
'''
if outputid == 'all':
#define number of rows
numsrc = self.SRC.shape[1]
nrows = np.divide(numsrc, ncols)
if np.remainder(numsrc, ncols) > 0:
nrows += 1
#prepare plots
fig, axes = plt.subplots(nrows=nrows,
ncols=ncols,
figsize=(40, 40),
facecolor='white')
fig.subplots_adjust(hspace=0.25, wspace=0.2)
i = 0
for ax in axes.flat:
if i < numsrc:
ax = plotbar(ax,
self.SRC[:, i],
self._namelist,
width=width,
addval=addval,
sortit=True,
*args,
**kwargs)
ax.set_ylabel('SRC', fontsize=15)
ax.yaxis.label.set_rotation('horizontal')
ax.yaxis.set_label_coords(-0.02, 1.)
if outputnames:
ax.set_title(outputnames[i])
else:
ax.set_axis_off()
i += 1
else:
print 'SRC values of output ', outputid, ' is shown in graph'
fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1 = plotbar(ax1,
self.SRC[:, outputid],
self._namelist,
width=width,
addval=addval,
sortit=True,
*args,
**kwargs)
ax1.set_ylabel('SRC', fontsize=20)
ax1.yaxis.label.set_rotation('horizontal')
ax1.yaxis.set_label_coords(-0.02, 1.)
plt.draw()
return fig, ax1